The Welfare E ects of Vertical Integration in Multichannel ... · The Welfare E ects of Vertical...

The Welfare Effects of Vertical Integration

in Multichannel Television Markets∗

Gregory S. Crawford† Robin S. Lee‡ Michael D. Whinston§ Ali Yurukoglu¶

PRELIMINARY

January 2, 2015

Abstract

We investigate the welfare effects of vertical integration of regional sports networks (RSNs)in U.S. multichannel television markets. Vertical integration can enhance efficiency by aligninginvestment incentives and/or reducing double marginalization, but can also harm welfare dueto foreclosure and raising rivals’ costs incentives. We measure these competing effects in thecarriage, channel placement, and pricing decisions of regional sports networks (RSNs) by affil-iated and unaffiliated cable and satellite television distributors. We first carry out descriptiveanalyses that compare integrated and non-integrated RSNs and distributors’ prices and car-riage, and viewership ratings. We then estimate a model of viewership, subscription, distributorpricing, and input cost bargaining. We use the estimated model to analyze the welfare effectsof simulated vertical mergers and de-mergers, and the “terrestrial loophole” introduced in the1992 Cable Act by the U.S. Federal Communications Commission.

1 Introduction

The welfare effects of vertical integration is an important, but controversial, issue. The theoretical

literature on the pro- and anti-competitive impacts of vertical integration is vast (c.f. Perry (1990);

Riordan (2008)), and typically contrasts potential efficiencies related to the elimination of double

marginalization (Spengler (1950)) and the alignment of investment incentives (Willamson (1985);

Grossman and Hart (1986)) with the potential for losses arising from incentives to foreclose rivals

and raise their costs (Salop and Scheffman (1983); Krattenmaker and Salop (1986); Hart and Tirole

(1990); Ordover et al. (1990)). However, despite a growing literature, empirical evidence on the

quantitative magnitudes of these potential effects (and the overall net impact) is still limited.

This paper attempts to quantify the welfare effects of vertical integration in cable and satellite

television in the context of high value sports programming in the US. Whether or not the ownership

∗We would like to thank ESRC Grant RES-062-23-2586 (Crawford), the NYU Stern Center for Global Economyand Business (Lee), and the Toulouse Network for Information Technology and the NSF (Whinston) for support. Allerrors are our own.†Department of Economics, University of Zurich; [email protected]‡Department of Economics, Harvard University; [email protected]§Department of Economics and Sloan School of Management, MIT; [email protected]¶Graduate School of Business, Stanford University; yurukoglu [email protected]

1

of content by distributors harms welfare is at the heart of several recently proposed (e.g., Comcast

and Time Warner, AT&T and DirecTV) and consummated (e.g., Comcast and NBC, approved in

2011) mergers in the television industry. The attention these mergers attracted is due to the indus-

try’s overwhelming reach: nearly 90% of the 115 million television households in the US subscribe

to multichannel television, and the mean individual consumes about three hours of television per

day. Regional sports programming alone comprises $4.1 billion per year in negotiated input fees

paid by distributors to content providers, with an additional $700 million in advertising dollars

spent on these channels.

Our focus on the multichannel television industry, and in particular sports programming, is

also driven by several factors that create empirical leverage to address this question. First, there

is significant variation across the industry in terms of ownership of content by multichannel video

programming distributors (MVPDs). Second, although this variation is primarily at the national

level for most channels, regional sports networks (RSNs) are present in smaller geographic areas,

and thus provides useful variation in ownership patterns both across regions and over time. Third,

the industry is the subject of significant regulatory and antitrust attention in addition to merger

review, including the application of “program access rules” and exceptions to this rule, such as

the “terrestrial loophole” which exempted certain distributors from supplying integrated content

to rivals.

The heart of our paper is the specification and estimation of a structural model of the multi-

channel television industry that captures consumer viewership and subscription decisions, MVPD

pricing and carriage decisions, and bargaining between MVPDs and content providers. Our paper

builds off and significantly extends the model in Crawford and Yurukoglu (2012) into an empirical

framework for the analysis of vertical integration and mergers, and importantly incorporates: (i)

incentives to foreclose rivals’ access to inputs, (ii) the potential for double marginalization, and

(iii) the possibility of imperfect coordination and internalization within an integrated firm. We use

data on both aggregate and individual level consumer viewership and subscription patterns with

price, quantity, and channel carriage for cable and satellite at the local market level over the years

2000 to 2010.

We leverage the structural model in this paper to highlight the mechanisms through which

pro-competitive and anti-competitive effects of vertical integration might occur. In particular, we

estimate: (i) the degree to which firms internalize the profits of integrated units when making

pricing and bundling decisions; and (ii) the incentive for an integrated channel to deny access to

rival distributors. Central to identifying both of these effects are our estimates of the changes

in firm profits from the addition or removal of an RSN from its bundles, which in turn relies on

using variation in distributor market shares as channel bundle offerings change to inform how much

consumers value content when subscribing to an MVPD, and variation in observed viewership

patterns and negotiated input fees across channels to infer the relative values consumers place

on different channels. Given these estimated profit effects, the pro-competitive effect of vertical

integration is identified from the degree to which carriage is higher for integrated distributors, while

2

the anti-competitive effect is identified by lower supply to downstream rivals of integrated channels.

We use our estimated model to conduct policy counterfactuals which examine the role of pro-

gram access rules and the “terrestrial loophole,” which was introduced in the 1992 FCC Cable

Act. Our findings indicate that integrated firms imperfectly internalize incentives to reduce double

marginalization, and that regulations prohibiting exclusive dealing by integrated firms have been

effective in many markets. By closing the “loophole” in certain markets, as the FCC voted to do

in 2010, and providing access to satellite distributors of those RSNs that were foreclosed to them,

we find that satellite penetration would increase by 20% and consumer welfare would increase by

$31 million annually in aggregate in the affected markets of Philadelphia and San Diego. On the

other hand, allowing integrated cable distributors to potentially foreclose satellite from carriage of

integrated RSNs in other markets not subject to the loophole—potentially allowed by the sunset-

ting of the Program Access Rules in 2012—would lead to changes in two large markets: Chicago

and the San Francisco Bay Area, decreasing satellite market shares by 10%, and resulting in an

$60 million annual aggregate consumer welfare loss.

At the moment, our analysis is partial. Ongoing work will examine price effects in markets

where rivals are supplied by the integrated firm, and the overall welfare effects of vertical integration

weighing reductions in double marginalization against foreclosure and raising rivals’ costs effects.

Related Literature. Previous work in the cable industry, including Waterman and Weiss (1996),

Chipty (2001), and Chen and Waterman (2007), have primarily relied on reduced form cross-

sectional analyses for a limited subset of channels and found that integrated cable systems are

more likely to carry their own as opposed to rival content. One exception that uses variation

over time is Suzuki (2009), which studies the 1996 merger between Time Warner and Turner

broadcasting and finds that integrated channels were more likely to be carried by Time Warner

systems following the merger, and that non-integrated rival channels were less likely to be carried

in Time Warner markets after the merger. In this regards, both this and our companion paper

(Crawford et al., 2014) complement previous work in the cable industry with a richer panel dataset,

and with a structural model we are able to both provide welfare measurements and shed light on

the mechanisms through which vertical integration has an effect on welfare.

This paper also adds to the growing empirical literature on the effects of vertical integration

or arrangements (e.g., Asker (2004), Hastings and Gilbert (2005), Hortacsu and Syverson (2007),

Villas-Boas (2007), Houde (2012), Lee (2013)). We build on existing approaches by explicitly

accounting for both efficiency and foreclosure incentives.1 We also allow for the potential imperfect

internalization of incentives across divisions within an integrated firm. Furthermore, our framework

allows for potential changes in certain non-price product characteristics (such as bundle offerings)

due to integration.

1See also Conlon and Mortimer (2013).

3

2 Institutional Detail and Data

Our study analyzes the US cable and satellite industry for the years 2000 to 2010, with a special

focus on “Regional Sports Networks” (RSNs). We briefly describe the industry and RSNs. We

then describe the data that we use to estimate the model.

Unless otherwise noted, the tables and figures referenced in this section are contained in Ap-

pendix B.

2.1 Vertical Affiliation in Multichannel Television Markets

In the time period we study, the vast majority of households in the US were able to subscribe to a

multichannel television bundle from one of three downstream distributors: a local cable company

(e.g., Comcast, Time Warner Cable, or Cablevision) or one of two nationwide satellite companies

(DirecTV and Dish Network). Cable companies transmit their video signals through a physical

wire whereas satellite companies distributed video wirelessly through a south-facing satellite dish

attached to a household’s dwelling. The majority of distributors’ revenue comes from subscription

to three different bundles of programming: a limited basic bundle which retransmits over-the-

air broadcast stations, an expanded basic bundle containing 40-60 of the most popular channels

available on cable (e.g., AMC, CNN, Comedy Central, ESPN, MTV, etc.), and a digital bundle

containing between 10 to 50 more, smaller, niche channels.

Downstream distributors negotiate with content producers over the terms at which the distrib-

utors can offer the content producers’ channels to consumers. These negotiations usually center

on a monthly per-subscriber (“affiliate”) fee that the downstream distributor pays the channel for

every subscriber who has access to the channel, whether the subscriber watches it or not.

This study focuses on the effects of vertical integration between distributors and Regional Sports

Networks. RSNs carry professional and college sports programming in a particular geographic re-

gion. For example, the New England Sports Network (NESN) carries televised games of the Boston

Red Sox and the Boston Bruins that aren’t concurrently being televised nationally. Metropolitan

areas can have multiple RSNs. For example, in the New York City metropolitan area, there are four

different RSNs (Madison Square Garden (MSG), MSG Plus, SportsNet NY, and Yankees Enter-

tainment and Sports (YES)). Some RSNs also serve multiple metropolitan areas. For example, the

Sun Sports network holds the rights to the Miami Heat and the Tampa Bay Rays, amongst others.

Most RSNs are vertically integrated with a downstream distributor. Table 6 provides a variety of

information about the largest RSNs in the US, including their availability, their average (across

systems and years) affiliate fee, and average (across DMAs and years) viewership. Figure 8 shows

each RSN’s years of operation between 2000 and 2010 and ownership affiliation with a downstream

distributor.2 According to industry estimates, RSNs command the second-highest per-subscriber

2DirecTV, the largest satellite operator and second-largest US distributor of multichannel television, was until2009 owned by Libery Media Corporation; though they are strictly separate companies, they share overlapping boardsof directors.

4

affiliate fees after ESPN.3 For example, NESN is reported to have per-subscriber monthly fees that

averaged $2.72 per month in 2010 whereas highly-rated national channels such as Fox News, TNT,

and USA hover around $1 per subscriber per month.

Regulatory policy towards RSN vertical integration There are several key features of the

regulatory environment for RSNs and vertically integrated content more generally that are relevant

during our sample period. First, during our sample period, vertically integrated firms were subject

to the “Program Access Rules” (PARs). These required that vertically integrated content be

made available to rival distributors at non-discriminatory prices, subject to final-offer arbitration

if required.

The Program Access Rules only applied, however, to content that was transmitted to the MVPD

via satellite. This covered all national cable channels (who need satellite transmission to cost-

effectively reach cable systems around the country) and most RSNs. A handful of RSNs, however,

transmitted their signal terrestrially (usually via microwave), thereby avoiding the jurisdiction of

the PARs. This was called the “terrestrial loophole” in the Program Access regulation. The most

relevant cases of the terrestrial loophole being used are Comcast SportsNet in Philadelphia and

SD4 in San Diego (owned by Cox Cable).4 Perhaps aa result of the terrestrial loophole, Major

League Baseball (MLB), National Basketball Association (NBA), and National Hockey League

(NHL) games in Philadelphia were only available through cable and not through DirecTV or Dish

Network. Similarly in San Diego, MLB games were only available through cable. This accident of

regulatory history will be an important source of identifying variation in our econometric estimation.

The Program Access Rules were introduced in 1992 and required renewal by the FCC every five

years. They were allowed to lapse in 2012 and replaced by rules giving the Commission the right

to review any programming agreement for anti-competitive effects on a case-by-case basis under

the “unfair acts” rules the Commission established in 2010 (FCC (2012)). The new case-by-case

rules explicitly include a (rebuttable) presumption that exclusive deals between RSNs and their

affiliated distributors are unfair.

2.2 Data

We collect a wide variety of data to analyze the effects of vertical integration. We have three

categories of data: (1) downstream prices, quantities, and characteristics of cable and satellite

bundles, (2) channel viewership data, and (3) channel input cost data. We briefly describe each in

turn.

3Affiliate fees are the fees paid by distributors to content providers for the ability to distribute the channel.4Time Warner Cable also employed the terrestrial loophole from 2006 to 2008 for the (then relatively new)

Charlotte Bobcats NBA franchise by placing some their games on News 14, a terrestrially delivered regional newschannel.

5

2.2.1 Downstream Prices, Quantities, and Characteristics

We combine data from multiple databases to construct downstream prices, quantities, and char-

acteristics. Our foundational dataset is the Nielsen FOCUS database. It provides, for each cable

system in the US, the set of channels offered, the number of homes passed, the total number of

subscribers (i.e. to any bundle), the owner of the system, and the zip codes served. We use the

years 2000 to 2010. We restrict our analysis to system-years in which the system faced no direct

wire-based competition.5 We combined these data with individual-level survey data household sur-

vey firms Mediamark and Simmons. Specifically, if a system-year had over 50 survey respondents,

we took the average of the market share from the FOCUS data and the cable market share among

the survey respondents. We further eliminate any system-year for which the subscriber data was

not updated from the previous year, or did not have at least 50 survey respondents. We use the

remaining system-years to construct our markets.

Each market is defined as a set of zip codes within a system-year served by a single cable

system and, by construction, both satellite providers. For cable systems, we aggregate over bundles

within a system, focusing on total system subscribers. Our demand model is therefore a distributor

choice model, rather than a bundle choice model.6 For satellite systems, we determine the number of

satellite subscribers separately for each of DirecTV and Dish Network using market shares estimated

from the individual-level survey data household survey firms Mediamark and Simmons. We use

Mediamark for 2000 to 2007, and Simmons for 2008-2010. We restrict our attention to markets

where we have at least 5 respondents in the individual-level data.7 Furthermore, we gathered

historical channel offerings and prices for DirecTV and Dish Network through the Internet Archive

(archive.org).

We combined multiple sources of information on cable television prices. Systems regularly

post prices for their tiers of service on their websites and these websites are often saved in the

Internet Archive.8 We use the price of Expanded Basic Service, the most popular bundle chosen by

households and the bundle which typically contains all the channels in our analysis. Furthermore,

newspapers often report when prices change at local cable systems. Some newspapers report this

information every time cable prices change (typically yearly), providing valuable information about

the history of price changes for a single (often large) system or geographic family of systems owned

by the same provider. Finally, cable systems typically have “rate cards” describing their current

tiers, channels, and prices which they use for marketing or to inform customers of changes in these

offerings. These are sometimes stored online and can also sometimes be found. We searched the

5We do so because when a system faces competition from another cable operator we do not know the number ofsubscribers in the areas where the system faced competition relative to the areas where it didn’t.

6While we would prefer a bundle demand model, our subscriber data was not rich enough to estimate bundle-specific quantities. This isn’t overly limiting, as our focus is on the impact of vertical integration on inter-distributordemand.

7We dropped any constructed market whose total market share exceeded one or which had a zero market sharefor one of the satellite providers which happens naturally due to sampling error.

8Following industry practice, we refer to the set of channels offered at a given (incremental) price as a tier ofservice and the combination of tiers chosen by households as the bundle that they buy. Thus the expanded basicbundle (service) consists of the limited basic tier and the expanded basic tier.

6

Internet for all such information about cable prices and linked the information obtained to Focus

systems by hand based on the provider, principal community served, and other communities served

as reported in the newspaper or listed on the rate card. For system-years where we do not find a

price from websites, rate sheets, or newspapers, we link to the TNS Bill Harvesting database. These

data are individual-level bills for cable service which report the company providing the service, the

household’s expenditure, and their zip code. For a given system-year, we use the mean expenditure

for subscribers to that system. These data also provide the level of a tax on satellite television

service in states where it exists, which we use as an instrumental variable for price in demand

estimation.

Table 7 reports the average price, market share, and RSN, cable, and total channels offered

across markets and years in our estimation dataset. We use over 6,000 markets over 11 years, with

an average coverage of 31.5 million (roughly 30% of) US households.9 Average prices are quite

similar across providers, whether on an unweighted basis or weighted by the number of households

in the market. The satellite companies generally offer more channels on their Expanded Basic

service than the local cable system, but a similar number of RSNs.

2.2.2 Viewership

We estimate demand using both bundle purchase and viewing data. We have two kinds of viewing

data: some at the level of individual households and others reporting aggregate viewing decisions

at the level of the Designated Market Area (DMA “ratings”).10 Average viewership for RSNs are

reported in Table 6 and average viewership for other cable networks are reported in Tables 8-9.

The first group of data come from our MRI and Simmons datasets described in the previous sub-

section. Our MRI data reports the number of hours watched for each of the sampled households

of 96 channels from 2000 to 2007 while our Simmons data reports the same information for 99

channels between 2008 and 2010. Our aggregate ratings data come from Nielsen. Reported is the

average rating on each of between 63 and 100 channels, of which 18 to 29 are RSNs, depending on

the year, in each of the 44 to 56 largest DMAs between 1998 and 2011.

Tables 6, 8, and 9 report summary statistics for our viewing data. Tables 8 and 9 report, for

each of our sources of viewing data, the mean rating for each of the 87 non-RSNs in either dataset,

as well as additional information from our household data. For example, the average rating for

the ABC Family Channel in the Nielsen data across the 747 DMA-years for which the information

was recorded is 0.418. This is measured in percentage points, so it suggests a household selected

at random in one of these years and DMAs would be watching the ABC Family Channel with

probability 0.418 percent. While small, this is above average for cable networks. Similarly, the

average rating for the RSN, Yankees Entertainment & Sports (YES) from Table 6, is 0.27. For

9While we observe the population of channel lineups, incomplete reporting of subscriber information in the Focusdataset and the inability to collect cable prices in some markets prevents us from constructing the information weneed in every US cable market.

10DMAs are mutually exclusive and exhaustive definitions of television markets created by Nielsen and used forthe purchase of advertising time.

7

RSN viewership, we have additional information about the average RSN rating by platform chosen

by households (i.e. cable or each satellite operator), which we report there.

Our household-level data provide further details about viewing which are summarized in the

remaining columns of Tables 8 and 9. The last column reports the share of households on average

across DMAs and years that report any viewing of that channel. As noted in Crawford and

Yurukoglu (2012), this provides valuable information about whether a household has any interest

in a channel that we will use to inform the estimated distribution of preferences for channels across

households.

2.2.3 Average Affiliate Fees

As described earlier, affiliate fees are the monthly per-subscriber charges paid by distributors to

content providers for the ability to distribute the channel. SNL Kagan maintains a database

with aggregate information about individual cable television networks, both nationally-distributed

networks like CNN and ESPN as well as RSNs like the family of Comcast and Fox networks. For

many networks, we use information about the average affiliate fee paid by cable systems to each

such network. For cable channels, we have information about affiliate fees paid by between 120 and

210 channels per year between 1998 and 2011. For RSNs, we also have information about the total

national subscribers served by each of 88 providers between 1998 and 2011. These are also reported

in Tables 6, 8, and 9. The average affiliate fee in our data is $0.16 per subscriber per month for a

nationally distributed channel and $1.45 for an RSN.

3 Model

In this section, we develop an industry model that predicts: (i) household viewership of channels;

(ii) household demand for multichannel television services; (iii) the prices and bundles that are

offered by distributors; and (iv) the negotiated distributor-channel specific input costs.

Index consumer households by i, markets by m, and time periods by t. There are a set of

“downstream” multichannel video programming distributors (MVPDs) Ft and “upstream” channels

Ct active in each period. MVPDs create and maintain a distribution network and perform retail

activities such as billing, packaging, and technical support. Examples include Comcast, Time

Warner Cable, Cox, Cablevision, RCN, DirecTV, Verizon FioS, AT&T U-Verse, and municipal

cable companies.

Let the set of MVPDs active in a given market-period be denoted Fmt. We will assume that

each distributor f ∈ Fmt in each period offers a single “bundle” in market m, where a household

subscribing to this bundle pays a price pfmt and has access to a set of channels Bfmt ⊆ Ct.11

We assume the following timing: in stage 1 channels and distributors bargain bilaterally to

decide input costs, and distributors also simultaneously set prices and make carriage decisions for

each market in which they operate; in stage 2 households in each market m choose which firm, if

11In the previous section we discussed how we deal with firms within a market offering multiple bundles.

8

any, to subscribe to; and in stage 3 households view television channels. We now provide details

of each stage and further assumptions, proceeding in reverse order of timing.

3.1 Stage 3: Household Viewing

Household i in market m and period t subscribing to firm j ∈ Fmt allocates its time between

watching available channels ({c} ⊆ Bj) and non-television activity (denoted by c = 0) to solve:

maxtij

vij(tij) =∑

c∈Bj∪{0}

γict1− νic

(tijc)1−νic (1)

s.t. : tijc ≥ 0 ∀ c

tijc = 0 ∀ c /∈ {Bj ∪ {0}}∑c∈Bj∪{0}

tijc ≤ T

Parameters γict and νic are household i’s taste parameters for channels, where γict sets the level

of marginal utility of household i from the first instant of watching channel c, and νic controls how

fast this marginal utility decays in the amount of time watched. We restrict νic to be equal, for a

given consumer, for all non-sport channels and the outside-option, and equal for all sports channels

(which include RSNs). We assume the distribution of each of the two components of ν (sports and

non-sports) are normal (truncated below zero and above one) with parameters Σν , and do not vary

over time.

We parameterize γit ≡ {γict}c, a Ct × 1 vector of household channel preferences, as:

γit ≡ χit · γ̃it

where χit is a vector whose components χict are Bernouilli random variables (i.e., 0 or 1) that

equals 0 with probability ρct, γ̃it is a vector where each component γ̃ict is drawn from an exponential

distribution with parameter σγct.

For RSNs, we scale γ̃ict by exp(−γbbict−γddic), where and bict ∈ [0, 1] represents the fraction of

teams carried on RSN c that are “blacked-out” (i.e., unable to have games televised in household

i’s market due to restrictions imposed by the team’s league) and dic is the average distance from

household i to the stadiums for the teams shown on RSN c (measured in thousands of miles).12

These terms allow for households to value an RSN differentially if the household cannot watch some

of the carried sport teams, or if the household lives further away from the carried teams’ stadiums.

3.2 Stage 2: Household Bundle Choice

Household i weighs the utility it would receive from watching the channels in the bundle offered by

each firm j against the price of the bundle and other characteristics to decide which firm, if any,

12We focus only on blackout restrictions for MLB, NBA, and NHL teams.

9

to subscribe to. We specify household i’s indirect utility conditional on subscribing to firm j as:

uijt = βvv∗ijt + βxxjt + βsatij + αpjt + ξjt + εijt (2)

where v∗ij is the indirect utility from the time allocation problem in (1), xjt are observable non-

price characteristics of bundle j such as year dummy variables and firm dummy variables, pjt is

the per-month subscription fee for bundle j, and ξjt is a scalar unobservable demand shock for

bundle j. Each consumer has a random preference for each satellite provider, βsatij , which is drawn

from an independent exponential distribution with parameter ρsatj ; βsatij = 0 if j is a cable bundle.

We assume that εijt is distributed Type I extreme value, that the outside option of no bundle is

normalized to ui0 = 0, and that each household chooses the bundle with the highest value of uij .

The probability that household i subscribes to bundle j in market m is:

sijmt =exp(βvv∗ijt + βxxjt + βsatij + αpjt + ξjt)

1 +∑

k∈Fmtexp(βvv∗ikt + βxxkt + βsatik + αpkt + ξkt)

(3)

where the market share of each bundle j (offered by firm f in market m at time t) is thus sjmt ≡∫i sijmtdHmt(i), where Hmt(i) is the joint distribution of household random coefficients (γ, ν, α) in

the market.

The demand for bundle j in market m is thus Djmt ≡ Nmtsjmt, where Nmt is the number of

television households in m.

3.3 Stage 1: Input Cost Bargaining, Distributor Pricing, and Bundling

In Stage 1, all distributors and channel conglomerates bargain over input prices {τfct}∀f,c, where

τfct represents the carriage fee that distributor f pays the owner of channel c for each of f ’s

household subscribers that receives channel c. Simultaneously, all distributors choose the prices and

composition of each of its bundles.13 That is, we assume that bargaining occurs simultaneously with

distributor pricing.14,15 We assume that both input prices, bundle prices, and bundle compositions

are optimal with respect to each other in equilibrium.

13A given distributor f often operates in many markets, and is choosing prices and bundle composition in each ofthese markets.

14See also Nocke and White (2007) and Draganska et al. (2010) who use a similar timing assumption. Formally, onecan think of separate agents of the distributor bargaining and making the pricing and bundle composition decisions.We leverage this assumption to simplify the computation and estimation of our model.

15An alternative timing assumption would be to assume that input prices are first negotiated, and then distributorprices and bundles are chosen. This would adjust firms perceptions of off-equilibrium actions: e.g., when bargaining,firms would anticipate different bundle prices to immediately be set if off-equilibrium input costs or disagreementwere realized. However, there may be reasons to believe that such a rapid response is unrealistic. Absent a fullyspecified dynamic model of firm bargaining and pricing, which is outside the scope of the current analysis, we believethe approach taken here to be a reasonable approximation.

10

3.3.1 Stage 1a. Distributor Pricing and Bundling

Every distributor f ∈ Ft chooses prices and bundles {pfmt,Bfmt}∀m:f∈Fmt to maximize its profits

given anticipated negotiated input fees τt ≡ {τfct}∀f,c. Profits for f across all markets are:

ΠMft ({Bmt}m, {pmt}m, τt;µ) =

∑m:f∈Fmt

ΠMfmt(Bmt,pmt, τt;µ)

where:

ΠMfmt(Bmt,pmt, τt;µ) = Dfmt

(pfmt −mcfmt

)+ µ

∑c∈Vft

ofct

( ∑g∈Fmt:c∈Bgmt

Dgmt(τgct + acmt))

(4)

We denote by Bmt ≡ {Bjmt}j∈Fmt and pmt ≡ {pjmt}j∈Fmt the set of bundles and associated prices

offered in the market, and by acmt the expected advertising revenue obtained by channel c per

subscriber to a bundle offering c.

The first component of an MVPD’s profit function in a given marketm, given by (4), is standard:

each bundle has a price and a marginal cost (mcfmt) that determine margins, and this is multipled

by demand. We assume that each MVPDs’ marginal cost can be decomposed into the sum of the

per-subscriber fees that f must pay to each channel c in the bundle, and a bundle-specific cost shock

that is the sum of non-channel related marginal costs, ωfmt: i.e., mcfmt ≡∑

c∈Bfmtτfct + ωfmt.

16

The second component of the profit function is non-standard, and represents the degree to which a

vertically integrated downstream unit values the profits that accrue to its upstream (i.e., channel)

units. These terms include per-subscriber fees and advertising revenues that accrue to integrated

upstream channels from its own viewers as well as from viewers of other distributors.17 The

parameter µ ∈ [0, 1] represents the extent to which a downstream MVPD f internalizes upstream

input fees and advertising revenues from all channels c ∈ Vft, where Vft represents the set of

channels owned by MVPD f in period t.18 The term ofct represents MVPD f ’s ownership share

of channel c at time t.19

In the absence of any frictions, µ would be equal to one; this would imply that the downstream

firm perfectly internalizes integrated upstream unit profits, and its strategic decisions maximize

total firm profit. Parameter µ could also be less than one, potentially representing divisionalization

that could arise from ignorance, poor management, optimal compensation under informational

frictions, or any other conflict between managers of different divisions within the same firm.

16Cost shocks include variable costs such as technical service labor costs or gas costs that are incurred on a per-subscriber basis.

17We omit portions of integrated channels’ profits which are not affected by f ’s pricing and carriage decisions, asthey do not affect the analysis.

18For our analysis, we only include in Vft the set of integrated RSNs. We will assume that c ∈ Vft (and hence, cis integrated with f) if MVPD f owns any percentage of channel c in period t.

19In the case that a third party has an x% stake in MVPD f and y% stake in channel c at time t, we assume thatofct = x%× y%.

11

Optimal Pricing and Bundling. We will leverage necessary conditions on the optimality of

MVPD pricing and bundling decisions in our estimation. Differentiating (4) with respect to pjmt

(and dividing by market size) yields the following pricing FOC:

∂ΠMfmt

∂pfmt=sfmt +

(pfmt −mcfmt

)∂sfmt∂pfmt

+∑g∈Fmt

( ∑c∈Bgmt∩Vft

µ× ofct(τgct + acmt))∂sgmt∂pfmt

(5)

In addition, we assume that the set of channels offered by each MVPD f in each market m satisfies:

Bfmt = arg maxBf

ΠMfmt({Bf , {Bgmt}g 6=f},pmt, τt;µ) (6)

Satellite Pricing and Bundling. If distributor f is a satellite MVPD (DirecTV or Dish), we

assume that the distributor sets a single national price and bundle. We assume that the bundle

offered by a satellite MVPD in any given market may differ from the national bundle only in the

set of RSN channels that are offered.

3.3.2 Stage 1b: Bargaining over Input Prices

Before describing how input fees are determined, we specify the profits each channel c contemplates

when bargaining with MVPD f in market m as:

ΠCcmt(Bmt,pmt, τt;µ, λR) =

∑g∈Fmt:c∈Bgmt

Dgmt

(τgct + acmt

+ µ× λR:fct

(ogct(pgmt −mcgmt) +

∑d∈Bgmt\c

oCcdt(τgdt + agdt)

))(7)

The first line reflects input fees and advertising revenues obtained from each bundle the channel is

available on; the second line incorporates potential profits of an integrated downstream MVPD, as

well as profits from other channels also owned by the same owner of channel c. We denote by oCcdt

the common ownership percentage of two channels c and d.

Both terms on the second line are multiplied by µ and λR:fct, where:

λR:fct =

1 if f and c are integrated (i.e., ofct > 0) ,

λR if f and c are not integrated .

We assume that λR:fct = 1 if c is owned by f and is bargaining with f ; this implies that a channel

and distributor that are integrated with each other place equal weight (given by µ) on each other’s

profits when bargaining with each other. However, if c is integrated but bargaining with a rival

distributor (i.e., the MVPD that c is bargaining with, f , is not an owner of c), then λR:fct = λR,

where λR ∈ [0, 1]; thus λR governs the extent to which an integrated upstream unit recognizes and

internalizes the effects of foreclosing the rival MVPD on the profits of its other integrated units.

12

c′ cOO

��f

ΠCcmt = πcmt + µ(πfmt + πc′mt)

(a) Internal Bargaining.

c′ c ^^

��g f

ΠCcmt = πcmt + µλR(πgmt + πc′mt)

(b) Bargaining with Rival MVPD.

Figure 1: Examples of ΠCcmt when c bargains with MVPD f .

In Figure 1, we provide an illustration of how channel c’s perceived profits when bargaining

with MVPD f may change depending on whether or not it is integrated with f . In Figure 1a, c is

integrated with MVPD f and another channel c′ (represented by the dashed square); in this case, c

will consider when bargaining with f its own profits (denoted by πcmt), consisting of input fees and

advertising revenues, as well as profits of its integrated distributor f and channel c′ (denoted by

πf and πc′) weighted by µ. We assume that πfmt includes f ’s subscription revenues net its costs;

profits πc′mt include c′’s input fees and advertising revenues. In Figure 1b, c is integrated with

another MVPD g and c′; in this case, c will consider when bargaining with f (a rival MVPD) its

own profits πcmt, and those of its integrated units πfmt and πc′mt weighted by µλR.

The parameter λR captures the extent to which an upstream unit has incentives to foreclose

access to the RSN to a rival distributor and lower the rivals’ bundle quality (thereby shifting demand

to the integrated distributor), an effect analogous to the “raising-rivals’-cost” effect discussed in

Salop and Scheffman (1983) and Krattenmaker and Salop (1986). We thus refer to λR as our “rival-

foreclosure” or “raising-rivals’-costs” (RRC) parameter. The indicator variable 1c∈Vgt (which equals

1 if c is owned by MVPD g) multiplies the margins accruing to a potentially integrated downstream

unit; the indicator variable 1∃h:c,d∈Vht (which equals 1 if channels c and d are both owned by the

same MVPD) captures input fees and advertising revenues to other channels owned by the same

parent MVPD.

We assume that, given channel c is carried on some of MVPD f ’s systems, the input cost

τfct between distributor f and channel c maximizes their respective bilateral Nash products given

the expected negotiated input costs of all other pairs and the expected prices and bundles for all

distributors:

τ̂fct(τ−fc,t,Bt,pt) = arg maxτfct

[∑m

[∆fcΠMfmt(Bmt,pmt, {τfct, τ−fc,t};µ)︸︷︷︸

GFTMfct

]

]ζfct(8)

×

[∑m

[∆fcΠCcmt(Bmt,pmt, {τfct, τ−fc,t};µ, λR)]︸︷︷︸

GFTCfct

]1−ζfct

13

where:

[∆fcΠMfmt(Bmt, ·)] ≡

(ΠMfmt(Bmt,pmt, {τ , τ−fc,t}; ·)−ΠM

fmt(Bmt \ fc,pmt, τ−fc,t; ·))

[∆fcΠCcmt(Bmt, ·)] ≡

(ΠCcmt(Bmt,pmt, {τ , τ−fc,t}; ·)−ΠC

cmt(Bmt \ fc,pmt, τ−fc,t; ·))

We denote by Bmt \ fc the set of all bundles Bmt in which we remove channel c from all bundles

offered by MVPD f . Thus, these terms represent the difference in either MVPD or channel profits

in market m if f no longer carried channel c. We will refer to GFTMfct and GFTCfct, which is the

sum of these terms across all markets, as the gains from trade (or bilateral surplus) for MVPD f

and channel c coming to an agreement.

This bargaining solution in which each pair of distributors and channels agree upon a set of input

fees which maximize the Nash product of their gains from trade is motivated by the model put forth

in Horn and Wolinsky (1988), and used by Crawford and Yurukoglu (2012) to model negotiations

between MVPDs and channel conglomerates.20 Each MVPD and conglomerate negotiate a single

input fee per channel that applies to all markets.

We can write the FOC of (8) for each channel c bargaining with MVPD f as:

ζfctGFTCfct(

∂GFTMfct∂τfct

) + (1− ζfct)GFTMfct(∂GFTCfct∂τfct

) = 0 (9)

where the derivative terms in (9) are:

∂GFTMfct∂τfct

=∑m

∂ΠMfmt

∂τfct= (−1 + (µ× ofct))

∑m∈Mfct

Dfmt

∂GFTCfct∂τfct

=∑m

∂ΠCcmt

∂τfct= (1− (µ× λR:fct × ofct))

∑m∈Mfct

Dfmt

and Mfct ≡ {m : c ∈ Bfmt} denotes the set of markets where c is on f ’s bundle. As we have

assumed that λR:fct = 1 whenever ofct > 0 (i.e., f and c are integrated and bargaining with one

another), it follows that ∂GFTMfct/∂τfct = −∂GFTCfct/∂τfct. We can thus re-write the FOC as:

GFTCfct = ΨfctGFTMfct (10)

where Ψfct ≡ (1− ζfct)/ζfct.This bargaining solution is not defined if µ× ofct = 1; under this case, f and c would perfectly

internalize each other’s profits when bargaining with one another, and the negotiated τfct would

be indeterminate. Also, in deriving our first-order condition, we are leveraging the assumption

that distributor bundle prices are set simultaneously with input costs, and there is no anticipated

20See also Grennan (2013), Gowrisankaran et al. (forthcoming), Ho and Lee (2013), and Collard-Wexler et al.(2014).

14

change in pfmt if τfct changes. Nonetheless, in equilibrium, both distributor pricing FOCs given by

(5) and input cost bargaining FOCs given by (10) will hold for realized values of prices and input

costs.

Example. Consider the case in which MVPD f and channel c are both non-integrated entities

that bargain with one another in period t. The negotiated input fee τfct that satisfies the Nash

bargaining FOC given by (10) solves:

∑m∈Mfct

Dfmtτfct = (1 + Ψfct)−1

∑m∈Mfct

((Ψfct[∆fcDfmt](pfmt −mcfmt + τfct)

)(11)

−(Dfmtacmt +

∑g 6=f :c∈Bgmt

[∆fcDgmt](τgct + acmt

)))

The LHS of this expression is the total payment made by f to c, and the RHS is a fraction of

the gains from trade due to agreement: the first term on the RHS represents f ’s increased profits

(net of payments to c) due to more subscribers induced by the carriage of channel c, and the

remaining terms represents (the negative of) c’s gains from being carried on f . Intuitively, the

more f gains from the relationship, the higher the total payment that is made; the more c gains

from the relationship, the lower the total payment. If f and c’s Nash bargaining parameters were

equal, then (1 + Ψfct)−1 = 1/2 and these gains from trade would be split in half.

The Role of λR. In our model, λR captures the internalization of an integrated downstream

MVPD’s profits when an integrated channel bargains with another distributor. Consider channel c

owned by MVPD f bargaining with rival distributor g (e.g., a satellite distributor). When λR > 0,

c’s desire to the increase downstream profits of f lowers c’s gains from trade when bargaining with

the non-integrated rival distributor g compared to when λR = 0. This may lead to the elimination

of overall gains from trade, and can result in non-supply of c to g. However, even if there are still

positive gains from trade, since these gains will be lower for c when λR > 0, the bargaining process

will lead to an increased input fee (τgct) for the rival distributor. Thus, even if g is still supplied

with channel c, its costs are raised; in equilibrium, this can lead the rival to increase the price of

its bundles to consumers.

4 Estimation and Identification

We estimate our model in two main stages.

In the first stage, we estimate θ ≡ {θ1,θ2,θ3}, where:

1. θ1 ≡ {Σν ,ρ,Σγ , γd, γb}, where Σγ ≡ {σγct}∀c and ρ ≡ {ρct}∀c,t, determines household viewer-

ship decisions. The first parameter of θ1, Σν , governs the distribution of household “decay”

15

parameters, and the remaining parameters govern the distribution of γ (household tastes for

channels).

2. θ2 ≡ {βv,βx,ρsat, α}, where ρsat ≡ {ρsatDirecTV , ρsatDish}, determines household bundle choice.

3. θ3 ≡ {µ} represents the extent to which integrated conglomerates and distributors internalize

profits across upstream and downstream units when pricing, bargaining, and choosing other

strategic variables.

Initially, we assume that Ψfct = 1/2 ∀f, c, t, and that distributors and channels have the same Nash

bargaining parameters.

In the second stage, we estimate our RRC parameter, λR.

Program Access Rules. To partially capture the impact of program access rules, we will assume

that λR = 0 in non-loophole markets and estimate our first stage parameters using only these

markets. We then estimate λR using only the markets in our data in which the terrestrial loophole

was used by RSNs (i.e., Philadelphia and San Diego).

4.1 First Stage Estimation

4.1.1 Moments used in Estimation

We estimate the model parameters via GMM, using the following moments derived from the model

described in the previous section.

Household Viewership. We use the difference between the following viewership moments ob-

served in the data and predicted by the model:

1. Summing across markets, the mean viewership for each channel-year;

2. Summing across markets, the number of households with zero viewership for each (non-RSN)

channel-year;

3. For each of five demographic groups, averaged across all RSN-years, the ratio of the mean

viewership of a given demographic group compared to the overall unconditional viewership

for an RSN channel-year.

Household Bundle Choice. Moments from the household bundle decision include:

1. We assume that each bundle’s unobservable characteristic is orthogonal to a vector of in-

struments: i.e., E[ξfmt(θ)Zξmt] = 0, where the expectation is taken across all markets, firms,

and years. For Zmt, we include bundle observable characteristics xfmt and predicted indirect

utility of channel viewing v∗fmt for the mean consumer; it also includes the satellite tax within

the market to instrument for pofmt. We recover ξfmt(θ) using the standard Berry et al. (1995)

inversion.

16

2. We match the covariance of cable vs. satellite subscription (0 or 1) with household income

by market, averaged across all markets and years.

Distributor Pricing, Bundling, and Bargaining. First, for any θ, the vector of input costs

{τfct} and bundle-specific marginal costs {mcfmt} can be directly computed using the pricing and

bargaining FOCs given by (5) and (10) (see Appendix for further details). We use these predicted

values of {mcfmt(θ)} and {τfct(θ)} in constructing the next set of moments:

1. Average Input Costs: the model’s predicted average input costs across MVPDs for each

channel should match observed average input costs:

Ef [τfct(θ)]− τ oct = 0 ∀c ∈ Ct .

We assume that any deviations reflect measurement error in τc, which are minimized. There

are Ct moments per year for estimation, which we construct by weighting estimated input

costs by MVPD market shares to approximate expectations across MVPDs.

2. Implied markups: the model’s predicted price-cost markups should match those observed

in the data (which we observe for Comcast, DirecTV, and Dish for 2007):21

Em[(pofmt −mcfmt(θ))/pofmt] = markupoft ∀f ∈ {Comcast,DirecTV,Dish} .

3. Bundle Optimality and Carriage: Equation (6) implies that every distributor f chooses

the optimal set of channels to include in each bundle in each market m. We will assume that

distributor f ’s true per-household profits in market m are given by π̃Mfmt(·), where:

π̃Mfmt(Bmt, ·) ≡ [πMfmt(Bmt, ·)− ν1fmt(Bfmt)] +

∑c∈B

ν2fct . (12)

and πMfmt(Bmt, ·) represents our (the econometrician’s) estimate of a firm’s per-household prof-

its. We introduce two types of disturbances in this definition: the first, ν1fmt(·), represents a

distributor-market-time specific disturbance which captures potential measurement or specifi-

cation error between our estimate of a firm’s profits and that observed by the firm; the second,

ν2fct, is a distributor-channel-time specific disturbance that is known to the distributor when

making its carriage decision (but realized subsequent to the bargaining stage), unobserved to

the econometrician, and may include non-measured per-household fixed incentives or costs of

carrying a channel.22

Now consider firm f and channel c that have negotiated an agreement: i.e., f carries c on

some bundles in some non-empty set of markets. For any m,m′ such that c ∈ Bfmt, c /∈ Bfm′t,21These values are [XX, YY, ZZ], which come from 2007 10K annual reports...22We assume that these disturbances enter linearly into both distributor and channel profits when bargaining so

that realizations of these disturbances do not influence the determination of equilibrium input costs.

17

a firm’s optimal bundling decision given by (6) implies that:

∆fc[πfmt(Bfmt, ·)]−∆fc[πfmt(Bfm′t ∪ fc, ·)] + (∆fc[ν1fm′t(Bm′t ∪ fc, ·)]−∆fc[ν

1fmt(Bmt)]) ≥ 0

where the ν2 disturbances cancel out.

We assume that {ν1fcmt(·)} is a mean-zero i.i.d. disturbance across firms, channels, markets,

and time. Then for each firm f and RSN c with agreement,

Em∈M+fct

[∆fc[πfmt(Bfmt, ·)]−∆fc[πf,m′(m),t(Bf,m′(m),t ∪ fc)]

]≥ 0 ∀f, c

Em∈M−fct

[−∆fc[πfmt(Bfmt ∪ fc), ·)] + ∆fc[πf,m′(m),t(Bf,m′(m),t)]

]≥ 0 ∀f, c

where M+fct denotes the set of markets in which f is active and carries channel c, M−fct

denotes those markets in which f is active but does not carry c, and m′(m) denotes a market

that has the opposite carriage decision for firm f and c (i.e., if c ∈ Bfmt, then c /∈ Bf,m′(m),t,

and vice versa) and is the closest (in Euclidean distance) to market m in terms of (weighted)

distance to the teams carried on RSN c and fraction of teams on c that are blacked out.23

These inequalities imply that the summed change in f ’s per-household profits in market m

and market m′ (where f carries c either in m or m′), when reversing its observed carriage

decisions in both markets and averaging across all markets m in which f either carries or

doesn’t carry c, is positive. If these inequalities did not hold, it would imply that f would

have a profitable deviation by changing its carriage decisions for c in certain markets.

These inequalities motivate minimizing the following moments in estimation:

1∑f∈Ft

|M+fct|

[ ∑f∈Ft

∑m∈M+

fct

∆fc[πfmt(Bfmt, ·)]−∆fc[πf,m′(m),t(Bf,m′(m),t ∪ fc, ·)]

]−

∀c

1∑f∈Ft

|M−fct|

[ ∑f∈Ft

∑m∈M−fct

−∆fc[πfmt(Bfmt ∪ fc, ·)] + ∆fc[πf,m′(m),t(Bf,m′(m),t, ·)]

]−

∀c

where [·]− ≡ min{·, 0} and | · | denotes the cardinality of the set. These sets of moments are

similar to those used in Crawford and Yurukoglu (2012) and follows Pakes et al. (forthcoming).

4.1.2 Identification

We now provide an informal discussion of how the parameters of the model are identified from

these moments.

The main parameters governing the distribution of γict (i.e., Σγ ,ρ) are primarily identified from

viewing behavior: e.g., channels watched more often have higher values of γict and lower values

23The rationale for this matching procedure is to match markets with comparable magnitudes of profitabilitychanges, and to be robust to the possibility that ν2fct might vary across very dissimilar markets.

18

of ρct. However, since we do not possess ratings for channels at the system level, we identify the

black-out and distance parameters γb, γd via the Bundle Optimality and Carriage moments; we

defer the discussion on this until the end of this subsection when discussing identification of µ.

Parameters governing household bundle choice, βx and βv, are identified from variation in

bundle market shares as observed bundle characteristics and channel utility changes: i.e., across

firms and years, and as channels are added and dropped from bundles. The satellite tax is an

instrument for price, and is used to identify the price sensitivity coefficient α. Information contained

in cable and satellite pricing margins helps identify the heterogeneity in preferences for satellite.

In particular, the relationship between satellite and cable market shares has strict implications for

predicted price elasticities (and hence implied markups) under a standard logit demand system

without preference heterogeneity; inclusion of a random preference for satellite (parameterized by

ρsat) assists with rationalizing observed markups for a given satellite market share.

In addition to observing how bundle market shares vary based on channel composition (which

has limited variation for some channels across markets), matching observed average input costs

negotiated for each channel {τ oct} to those predicted by the model {τfct(θ)} is crucial. First, our

model relates τfct(θ) to the gains from trade created when channel c contracts with firm f : i.e.,

differences in f and c’s profits (primarily realized from subscription and advertising revenues) when

f drops c. Thus, our model attempts to rationalize a channel with higher observed input costs τ oct

by predicting that this channel creates greater surplus from carriage: this is partly through the

term βvv∗ijt in a household’s bundle utility equation given by (2), which in turn is also a function

of parameters governing the distribution of γict, and how γict is scaled to enter into utility by νic

(which has a distribution paramaterized Σν)—i.e., a channel with a higher γic and lower decay

parameter νic than another will contribute more to a viewer’s utility from the same amount of time

the channel is watched.

To anchor this in an example, consider a single market and bundle with two channels c and

c′, and a single household i. Assume that viewers watch c′ twice as long as c. This could be

induced by many potential combinations of (γic, νic, γic′ , νic′); e.g., γic′ could be higher than γic

and νic = νic′ . If this were true, however, then c′ should obtain higher negotiated input costs as

it would be predicted to generate a higher surplus for a viewer, and hence there would be higher

gains from trade from carriage of c′ than c. However, if input costs are observed to be the same

for the two channels despite the difference in viewership, then the model would predict that the

rate of “decay” for channel c, νic, was in fact higher than νic′ (thereby allowing c to generate the

same utility for consumers—and hence same negotiated input costs—for a shorter amount of time

watched).

Now add to this example two additional markets: one market only has channel c available, and

another only has channel c′. If viewership patterns for these channels in the new markets were

similar to those in the first market, then variation in market shares for cable across these markets

as the channel composition of the bundles changed would inform the value of βv.

In a sense, the parameters governing the distributions of γic and νic help the model rationalize

19

01

23

4M

onth

ly In

put F

ees

0 .5 1 1.5Ratings

Non-Sports Sports

Sports and Non-SportsMonthly Input Fees vs Ratings

Figure 2: Negotiated monthly input fees and viewership ratings.

variation in both negotiated input costs and the market share of bundles as (both the mean and

variance of observed) viewership of channels changes across markets, controlling for channel carriage;

on the other hand, βv helps the model rationalize variation in market share of bundles as channel

carriage changes across markets, holding fixed patterns of viewership for these channels.

The reason that we allow for consumers to possess two different “decay” parameters νic for sports

and non-sports channels is motivated by the data, illustrated in Figure 2. Sports channels have

consistently higher negotiated input fees than non-sports channels with similar viewership patterns

(ratings), in cases receiving payments an order of a magnitude higher. Our model rationalizes

this by assigning a higher decay rate to sports channels, which predicts higher utility delivered to

consumers for a given amount of time the channel is watched; thus, sports channels are able to

negotiate higher input fees as they create greater gains-from-trade upon agreement with an MVPD.

Although the internalization parameter µ enters into the computation of several moments (in-

cluding any moment based off of recovered values of τfct(θ) and mcfmt(θ)), it will primarily be

identified off of the Bundle Optimality and Carriage moments. In particular, as µ increases, dis-

tributors have a greater incentive to carry an integrated channel for a fixed value τfct(·); hence, the

model will help to rationalize higher carriage rates between integrated distributors and channels

(which is observed in the data). Similarly, we identify our black-out and distance parameters, γb, γd,

in a similar fashion. An example of the variation in the data that we leverage is illustrated in Fig-

ure 3: for both Comcast SportsNet Chicago and Comcast SportsNet Mid-Atlantic, non-carriage by

the integrated distributor (Comcast) is less likely than non-carriage by non-integrated distributors,

and non-carriage overall is more likely as the system is further from the RSN’s teams’ stadiums and

when teams on the RSN are blacked out (as in Michigan for CSN Chicago, and in Pennsylvania for

CSN Mid-Atlantic). Table 1 summarizes this relationship across all RSN’s and distributors in our

20

Figure 3: Carriage by integrated and non-integrated MVPDs of CSN Chicago (left) and CSNMid-Atlantic (right)

sample: carriage of an RSN by a cable system is strongly increasing with the RSN and distributor

being integrated, and strongly decreasing in the distance between the system and the RSN’s teams’

stadiums and in the fraction of teams that are blacked out.

4.2 Second Stage Estimation

4.2.1 Recovery of λR

To recover our RRC parameter λR, we will use information provided by markets in which distrib-

utors are able to exclude competitors from carrying an integrated RSN channel—i.e., terrestrial

loophole markets. The markets we focus on will be Philadelphia and San Diego, the channels in

question CSN Philadelphia (owned by Comcast) and 4SD (owned by Cox), and the competitors

excluded from carriage are satellite providers DirecTV and Dish.

To describe our approach, consider a channel c that is integrated with distributor f that is

“relevant” (i.e., offered and plausibly available to some set of distributors) in markets Mc. If we

observe that channel c does not contract with distributor g 6= f , we will assume that λR must have

been sufficiently large that c and g not contracting with one another is an equilibrium outcome. A

necessary condition for this is that there is no input fee τ̃gct such that c and g would both find it

21

Table 1: Regression of RSN Carriage on Integration Status and Distance

RSN Carriage Regression

Coeff. SE tIntegrated with RSN 0.143 0.026 5.46

Distance to RSN (mi) -0.001 0.000 -11.08N MLB Teams on RSN 0.070 0.019 3.62N NBA Teams on RSN 0.065 0.021 3.15N NHL Teams on RSN 0.210 0.028 7.49

RSN-Year FE YesMSO FE YesDMA FE Yes

R-squared = 0.5704N = 11063

Notes: Linear probability regression where the dependent variable is whether a system carries an RSN in a given

year. SE’s are clustered by RSN-year.

profitable to contract with one another:

∑m∈Mc

[(ΠMgmt({Bomt ∪ gc},pomt, {τ̃gct, τ̂−gct}; µ̂)︸︷︷︸

MVPD g’s profits in m with c

−ΠMgmt(Bomt,pomt, τ̂−gct; µ̂)

)(13)

+(

ΠCcmt({Bomt ∪ gc},pomt, τ̃gct; µ̂, λR)︸︷︷︸

c’s profits in m when supplying MVPD g

−ΠCcmt(Bomt,pomt, τ̂−gct; µ̂, λR)

)]≤ 0 ∀ τ̃gct

where the o superscript denotes observed variables, {Bomt ∪ gc} denotes the set of observed bundles

with the modification that g carries c on all of its bundles (in the relevant markets), ·̂ are estimated

values from the first-stage estimation, and τ̂−gct represents all input fees except those between g

and c.24 If (13) holds for all values of τ̃gct, the solution to the Nash Bargain between g and c given

by (8) is not defined.

Since we are evaluating a deviation in a model in which bundle composition, bundle prices, and

input fees are simultaneously determined, when computing “counterfactual” profits from agreement

between channel c and distributor g (the terms with underbraces in (13)), we will hold fixed bundle

prices and the channels carried when evaluating counterfactual profits upon carriage of c by g.25

Furthermore, in the case where g is a satellite distributor, it must carry c in all of the relevant

markets. In that case, condition (13) holds at any τ̃gct if and only if the joint profits of the two

24To be precise, input fees are not directly estimated; instead, we compute their implied values at the estimatedparameters θ̂: i.e., τ̂ ≡ τ (θ̂), where τ (·) is the solution to the Nash bargaining FOCs given by (10).

25The condition that there does not exist a deviation to carriage is not the same as testing whether carriage of cby g would comprise an equilibrium outcome, as this test would require (among other things) computing equilibriumprices and input fees conditional on carriage of c by g being known and anticipated by all firms in the market.

22

parties is larger with non-supply. We thus can test whether (13) holds for τ̃gct = 0 to determine

whether or not a deviation for c to supply g is profitable for both parties.

Multilateral Deviations. Alternatively, we may believe that it is not feasible for an integrated

channel c to be withheld by its cable owner f from one satellite provider, s, but provided to the

other satellite provider, s′. This can be motivated by regulation or legal constraints. In such a case,

the previous bilateral analysis may not be appropriate. Instead, we will determine whether, at the

observed set of bundles, input costs, and bundle prices, there are no gains from trade between c

and both satellite providers s and s′ (thereby ruling out the presence of this profitable deviation):

∑m∈Mc

[(ΠMsmt({Bomt ∪ {sc, s′c}},pomt, τ̃ ; µ̂)−ΠM

smt(Bomt,pomt, τ̂−{sct,s′ct}; µ̂))

(14)

+(

ΠMs′mt({Bomt ∪ {sc, s′c}},pomt, τ̃ ; µ̂)−ΠM

s′mt(Bomt,pomt, τ̂−{sct,s′ct}; µ̂))

+(

ΠCcmt({Bomt ∪ {sc, s′c}},pomt, τ̃ ; µ̂, λR)−ΠC

cmt(Bomt,pomt, τ̂−{sct,s′ct}; µ̂, λR))]≤ 0 ,

where the three lines represent s, s′, and c’s gains from trade from both s and s′ being supplied with

channel c, and τ̃ is equal to τ̂ except that τ̃sct = τ̃s′ct = 0. As in the case of bilateral deviations

before, we test whether or not (14) holds when the negotiated input fees between s and c and

between s′ and c equal 0.

Although bargains are happening simultaneously, we are assuming that both s and s′ believe

that if one of them was offered channel c, both satellite providers would be offered the channel.

This is consistent if all parties are aware of the constraints imposed under this scenario.26

We estimate a lower bound of λR, denoted λ̂R by finding the lowest value that ensures that either

(13) or (14) hold for all channel-distributor pairs that do not contract in the loophole markets.27

4.2.2 Recovery of ν2

Although not necessary to interpret our main model estimates, the recovery of unobserved firm-

channel specific carriage disturbances ν2fc will be useful to test the robustness of our analyses.

Again, consider a given distributor f with an agreement with channel c. Taking expectations

over (6) implies that:

−Em∈M+fct

[∆fc[πfmt(Bfmt, ·)]

]≤ ν2

fct ≤ −Em∈M−fct[∆fc[πfmt(Bfmt ∪ fc, ·)]

]∀f, c

26This is similar to assuming “symmetry beliefs” (McAfee and Schwartz, 1994) in that satellite distributors antic-ipate that either both are supplied or both are excluded from c when receiving an off-equilibrium offer; however, wedo not impose the restriction that satellite distributors anticipate receiving the same input fee for carriage. In ourmodel, since prices and carriage are simultaneously determined, the input fee offered to s′ does not affect s’s profitsand hence its decision to .

27For now, we will assume away the specification error introduced in (12): i.e., ν2gct = 0.

23

Table 2: Estimates of Key Parameters

Parameter Estimate SE

νsports 0.60νnon−sports 0.94

γd (Distance Decay) -6.034

α0 -0.127βv 0.016

βsatDirecTV 11.594βsatDish 15.886ρsat 0.139

µ 0.90λr 0.79

Notes: Key parameters from the first and second stage estimation of the full model.

which provides a set of bounds for ν2fct.

We will assume a parametric distribution over ν2 ≡ {ν2fct}fct: i.e., ν2

fct ∼ N(υ1, υ2), and

estimate υ ≡ {υ1, υ2} using MLE (conditional on all the bounds being non-empty).

5 Results

Estimates of the key parameters of our model are reported in Table 2. We discuss our estimates

primarily through how they influence predicted moments relating to (i) viewership patterns, (ii)

consumer bundle choices and implied firm pricing decisions, (iii) negotiated input fees, and (iv)

carriage and bundling decisions.

5.1 Channel Valuations

Our model predicts the willingness-to-pay (WTP) for each channel by household by computing the

contribution of a given channel to bundle utility (v∗ijt in (2)), and multiplying it by our estimates

of βv/αi to convert it into dollars.

The distribution of household WTP for 8 national channels is provided in Figure 4a. In Ap-

pendix B, Table 10 reports WTP estimates for all national channels and Table 11 reports WTP

estimates for the RSNs.

Our estimate of the RSN distance-decay is negative, and implies that consumers derive less

utility from watching a RSN the further they are from the stadium of the main team carried by

that RSN: a household 100 miles away from a channel’s main team stadium values that channel

only 55% (exp(−6.034 × 0.1)) as much as a household right next door to the stadium. Figure 4b

illustrates this pattern, and plots the predicted mean WTP of households for 4 different RSNs as

the distance from a household to an RSN’s team stadium increases.

Finally, we estimate different values of νsports and νnon−sports, where the higher value of νsports

implies that consumers’ marginal utility from watching sports channels falls faster than for non-

24

0 5 10 15 20 250

0.2

0.4

0.6

0.8

Monthly WTP in Dollars

Fra

ctio

n of

Con

sum

ers

American Movie Classics AMC

0 5 10 15 20 250

0.2

0.4

0.6

0.8


Fra

ctio

n of

Con

sum

ers

Discovery Channel

0 5 10 15 20 250

0.2

0.4

0.6

0.8


Fra

ctio

n of

Con

sum

ers

ESPN

0 5 10 15 20 250

0.2

0.4

0.6

0.8


Fra

ctio

n of

Con

sum

ers

Food Network

0 5 10 15 20 250

0.2

0.4

0.6

0.8


Fra

ctio

n of

Con

sum

ers

Fox News Channel

0 5 10 15 20 250

0.2

0.4

0.6

0.8


Fra

ctio

n of

Con

sum

ers

MTV

0 5 10 15 20 250

0.2

0.4

0.6

0.8


Fra

ctio

n of

Con

sum

ers

Turner Network TV TNT

0 5 10 15 20 250

0.2

0.4

0.6

0.8


Fra

ctio

n of

Con

sum

ers

USA

(a) Histograms of Monthly WTP.

0 0.1 0.2 0.3 0.40

2

4

6

8

10

12

Distance to Team Stadium (000s mi)

Mea

n W

TP

CSN Chicago

0 0.05 0.1 0.15 0.2 0.250

2

4

6

8

10

12

14

16


Mea

n W

TP

CSN Philadelphia

0 0.05 0.1 0.15 0.2 0.250

2

4

6

8

10

12

14


Mea

n W

TP

Fox Sports Ohio

0 0.1 0.2 0.3 0.40

1

2

3

4

5

6

7


Mea

n W

TP

MSG Network

(b) Mean WTP by Market-RSN as a function of distancefrom Market to RSN team stadiums.

Figure 4: Predicted WTP for channels.

02

46

8E

stim

ated

Mea

n W

TP

0 .5 1 1.5Estimated Ratings

Non-Sports Sports

Sports and Non-SportsEstimates: Monthly WTP vs Ratings

Figure 5: Estimated Monthly WTP vs Ratings for Sports and Non-Sports Channels

sports channels; in turn, this implies that consumers derive higher utility from sports channels than

non-sports channels for the same amount of time spent watching each. Our model thus predicts that

sports channels receive higher negotiated input fees for the same viewership ratings, as depicted in

Figure 5.

25

5.2 Pricing and Bundle Choices

In Table 3, we report average predicted own and cross price elasticities and implied margins for

cable and satellite MVPDs predicted by our model. Demand for the average cable system (-1.6) is

Table 3: Elasticities and Margins

Elasticity of row with respect to price of column: Cable DirecTV Dish

Cable -1.624 0.280 0.183DirecTV 1.895 -2.629 0.111

Dish 2.603 0.189 -3.593

Mean Cable Margin 0.765Mean DirecTV Margin 0.539

Mean Dish Margin 0.451

OLS Logit Price Coefficient -0.0046** (t: -2.40)IV Logit Price Coefficient -0.0987*** (t: -6.17)

Notes: This table reports mean price elasticities and margins by cable/DirecTV/Dish, as well as the effect of thesatellite tax instrument on the price coefficient in a logit demand system.

more inelastic than for satellite (-2.6 and -3.6), which is consistent with its larger market shares and

higher predicted margins. The margins implied by the model are close to the observed Comcast

(.67), DirecTV (.57), and Dish (.49) margins computed from each company’s 2007 annual report.

In addition, the bottom panel of Table 3 reports the effect of instrumenting for bundle prices

using the satellite tax instrument that was discussed in the previous section. In a logit demand

system, instrumenting for price yields a 20 times larger estimated price coefficient, consistent with

the presence of a positive correlation between price changes and unobservable bundle characteristics.

5.3 Internalization and RRC Parameters.

We now turn to the estimates and magnitudes of µ and λR.

Our estimated value of µ indicates that firms do internalize the profits of other integrated units

when making decisions: i.e., when pricing and determining carriage on its bundles, an MVPD inter-

nalizes potential effects on input fees and advertising revenues accruing to integrated channels; and

when bargaining internally, an integrated MVPD and channel face reduced double marginalization

incentives. Insofar our estimated value of µ < 1, however, such internalization is imperfect.

Our estimated lower bound for λR is .79, which indicates also a high level of internalization

when an integrated channel bargains with rival MVPDs. Figure 6 graphs the total three party

surplus between the integrated channel and the two satellite providers in the two loophole markets

we examine (Philadelphia and San Diego). We see that for values of λR lower than .5, it is not

an equilibrium for either channel to exclude both satellite providers as there would be a profitable

deviation (for some negotiated set of input fees) for the channel to be supplied. However, for

values of λR between approximately .5 and .79, we can rationalize exclusion in San Diego but not

26

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.3

−0.2

−0.1

0

0.1

0.2

0.3

0.4RSN’s and Exclusion

λr

Thr

ee P

arty

Sur

plus

4SD San Diego − CoxCSN Philadelphia − Comcast

Figure 6: Three Party Surplus as a function of λr in Philadelphia and San Diego.

Philadelphia. Only for values of λR ≥ .79 does our model rationalize exclusion in both of these

loophole markets.

6 The Welfare Effects of Vertical Integration

In this section, we use estimates from our model to perform counterfactual exercises that illustrate

how vertical integration affects input cost negotiations, distributors’ pricing and carriage decisions,

and—ultimately—firm and consumer welfare.

Focusing on the year 2007, we perform two counterfactuals. In our first counterfactual, we close

the “terrestrial loophole” in Philadelphia and San Diego. In our second counterfactual, we explore

the impact of relaxing program access rules in markets where the “terrestrial loophole” did not

apply: we focus on the negotiations between integrated RSNs and rival MVPDs, and examine the

extent to which the RSN would wish to deny access to other distributors. In both counterfactuals,

we quantify the welfare impact of this change in carriage.

Future exercises (currently in progress) will control for adjustments in negotiated input fees

and prices following a counterfactual change, and explore predicting market outcomes if MVPDs

had to divest integrated RSNs (which is equivalent in our model to assuming that upstream and

downstream units of integrated firms do not internalize each others incentives when making strategic

decisions: i.e., setting µ = 0).

Discussion. Before proceeding, it is useful to discuss the components of our model and the types

of responses permitted in these exercises in order to understand the range of effects that we allow

for when evaluating the impact of vertical integration.

27

There are three primary supply-side decisions our model emphasizes: carriage decisions, negoti-

ations over input costs conditional on carriage, and bundle pricing. When an MVPD and a channel

are integrated, our estimated value for µ̂ > 0 implies that integrated downstream and upstream

units (at least partially) internalize joint profits when making all of these decisions.

For exposition, assume that MVPD f integrates with channel c, and there is a rival MVPD g

and another channel d. The following effects of vertical integration are admitted in our model:

I. When an integrated channel c bargains with a rival MVPD g (since λ̂R > 0), c internalizes

lost revenues to its integrated downstream MVPD f by supplying g (induced by making g a

more competitive rival and taking customers from f); c may thus have a greater incentive to

deny carriage to g than it would have if it were not integrated.

II. When an integrated MVPD f negotiates with other channels d, f internalizes viewership

changes to its integrated channel c when it carries channel d on its own bundles; if c and d

are substitutable, f may be willing to pay a lower negotiated input cost τfdt (and potentially

have less of an incentive to carry d), as the gains from trade from f carrying d are partially

mitigated by lost viewership and advertising revenues to c.

III. When an integrated MVPD f prices its bundles:

(a) f faces a lower perceived marginal cost as it internalizes payments made to c; hence,

double marginalization incentives induced by linear input costs may be mitigated;

(b) f internalizes input costs paid by rival MVPD g to integrated channel c, thereby alle-

viating bundle pricing pressure and increasing in its “effective” marginal cost as f now

partly benefits from customers lost to g (Chen, 2001).

The welfare effects of some of these incentives may be straightforward to sign ex ante; for others,

it is not clear. Effect I, for instance, may likely lead to consumer welfare losses: if g loses access

to c or pays a higher input price τgct, g’s subscribers may receive less utility from their bundle of

channels (from reduced choice or higher prices); f ’s prices may also increase in response.28 Effect

II, similarly, may have negative consumer welfare consequences if this leads to f (and g) increasing

prices. However, effect III has two components with potentially opposite effects: whereas IIIa would

favor lower bundle prices, IIIb serves to mitigate price competition and push prices higher.

At the moment, we focus on primarily the effects mentioned in I. In our current model, there are

many potential responses that we have not yet (or cannot) explicitly control for. Most importantly,

we have not modeled investment in channel and programming quality, which may increase upon

integration due to the alignment of incentives between upstream and downstream firms, and the

potential reduction of hold-up concerns surrounding counterparty specific investments.

28Upon g losing access to c, it is feasible that bundle prices may also fall due to either lower prices from g, or fromlower prices from f due to lower negotiated input costs from d. Furthermore, consumers re-optimizing their bundlechoice would also influence welfare predictions.

28

Table 4: Closing the Terrestrial Loophole in 2007

Market Share Surplus ($/month/capita)exc. w/o exc. change exc. w/o exc. change

4SD Integrated Cable: 0.739 0.716 -3.10% 13.271 12.864 -3.06%Cox Satellite: 0.106 0.132 23.80% 0.975 1.205 23.57%

Pop. 1052705 Consumer: 29.921 30.304 1.28%

CSN Phil. Integrated Cable: 0.646 0.612 -5.25% 10.915 10.421 -4.53%Comcast Satellite 0.159 0.199 25.19% 1.624 2.011 23.83%

Pop. 2762396 Consumer: 26.794 27.822 3.84%

Notes: This table presents changes in the model’s predicted market shares and surplus under a counterfactual scenariowhere the integrated RSN is forced to supply satellite distributors in Philadelphia and San Diego. The owner of thechannel in question and the population in each market is reported below the channel name. The columns “exc.” and“w/o exc.” indicate scenarios with exclusion and without exclusion of satellite. Surplus calcuations for integratedcable providers include surplus for the integrated RSN; all surplus figures are reported holding fixed existing inputfees and prices, assuming that the new input fee between satellite and the previously excluded RSN is 0, and in unitsof $ per month per capita

Consequently, we view our counterfactuals as being only partial equilibrium results (both in

their current as well as final form), and thus any interpretation of our findings must be made with

this in mind.

6.1 Exercise 1: Closing the Terrestrial Loophole (Disallowing Exclusion in Ex-

empted Markets)

Our first counterfactual simulates market outcomes in Philadelphia and San Diego when the inte-

grated RSN channels are forced to serve the two satellite providers.29 This would be equivalent to

assuming that λR < .59 in both markets so that exclusion would not be desirable on the part of

each RSN.

Results from this exercise are reported in Table 4. The two panels of the table report computed

market outcomes in San Diego and Philadelphia if 4SD and CSN Philadelphia were supplied by their

cable owners to both satellite providers Dish and DirecTV. Examining market shares, supplying

satellite providers with the RSN would increase their market share by 3-4 percentage points (25%);

almost all of this would be taken from cable, with less than a 1 percentage-point increase in total

MVPD market share in each market. In both markets, consumer surplus would be predicted

to increase by $0.40 and $1.00 per capita per month in SD and Philadelphia, corresponding to

approximately $5M and $34M per year, respectively.

29These two markets were exempt from the FCC’s Program Access Rules by the “terrestrial loophole” as theirdelivery system did not rely on satellite transmission.

29

6.2 Exercise 2: Removing Program Access Rules (Allowing Exclusion in All

Markets)

Our second counterfactual simulates the market outcomes when MVPDs with integrated RSN

channels are able to potentially exclude satellite distributors from carriage. As we have assumed

that λR = 0 in these markets during estimation, we will here examine the negotiations between

the integrated cable provider and satellite in these markets when λR > 0. In particular, having

identified an estimate of the lower bound of λ̂R, we will explore how carriage decisions will change

for values of λR > λ̂R.

We focus on RSN’s which are vertically integrated with cable.30 These RSN’s are: Comcast

SportsNet Bay Area, Comcast SportsNet California, Comcast SportsNet Chicago, Comcast Sport-

sNet Mid-Atlantic, Comcast SportsNet Northeast, Comcast SportsNet Northwest, Comcast Charter

Sports Southeast, Cox Sports TV (New Orleans), Madison Square Garden (integrated with Cable-

vision), Madison Square Garden Plus (integrated with Cablevision), and SportsNet NY (integrated

with Time Warner Cable and Comcast).

For an integrated RSN channel c relevant in markets Mc carried by satellite provider s, and

for a given λR > λ̂R, we will determine whether or not at the observed set of bundles, input costs,

and bundle prices, there are no GFTs between c and both satellite providers s and s′:

∑m∈Mc

[(ΠMsmt(Bomt,pomt, τ̂ ; µ̂)−ΠM

smt({Bomt \ {sc, s′c}},pomt, τ̂ ; µ̂))

+(

ΠMs′mt(Bomt,pomt, τ̂ ; µ̂)−ΠM

s′mt({Bomt \ {sc, s′c}},pomt, τ̂ ; µ̂))

+(

ΠCcmt(Bomt,pomt, τ̂ ; µ̂, λR)−ΠC

cmt({Bomt \ {sc, s′c}},pomt, τ̂ ; µ̂, λR))]

< 0 . (15)

where (15) differs from (14) in that here, the integrated channel is removed from the observed

bundles in the data (as we are examining markets exempt from the loophole and potentially subject

to PARs) as opposed being added.

Results. We report our findings in Table 5 for λR = .79 and λR = 1. We find that even when

λR = 1, exclusion is not predicted in approximately half of our markets with an RSN owned by

a cable MVPD (5 out of 11).31 Figure 7 plots the three party surplus (integrated cable owner

and the two satellite providers) for 6 cable-integrated RSNs for different values of λR; these gains

are negative for 3 of the channels, thus implying that expanding the terrestrial loophole to other

markets would potentially lead to exclusion by an integrated cable provider.

The main drivers that make exclusion less likely for a cable-integrated RSN is a larger mar-

ket share of satellite and smaller coverage of the integrated cable provider; i.e., the larger is the

30We also check the Root Sports networks (Northwest, Pittsburgh, and Rocky Mountain) which are integratedwith DirecTV. However, because cable’s market share is so large, our estimated model predicts that DirecTV wouldnot profit from refusing to provide Root Sports to cable.

313 other markets listed, in which exclusion is not predicted, are for RSNs channels owned by DirecTV.

30

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.2

−0.1

0

0.1

0.2

0.3

0.4

0.5Comcast RSN’s and Exclusion

λr

Thr

ee P

arty

Sur

plus

Bay AreaCaliforniaChicagoMid−AtlanticNew EnglandNorthwest

Figure 7: Three Party Surplus as a function of λr in for currently protected RSN’s.

satellite share, the more that an RSN integrated with a cable provider would lose from excluding

satellite in terms of foregone input fees and potential ad revenues; the smaller the integrated cable

provider’s share, the smaller would be the loss in subscription revenue borne by the RSN’s inte-

grated downstream distributor when consumers substitute to a satellite MVPD (due to satellite’s

lower margins).

These findings indicate that some channels owned by a cable provider would have an incentive

to deny access to satellite if program-access rules were relaxed. To compute predicted welfare

outcomes from this counterfactual, we examine each RSN channel and market in isolation, and

hold fixed: (i) negotiated input prices and carriage decisions for all other channels; and (ii) bundle

prices for all distributors in these markets. However, it is important to stress that if “opening up”

the terrestrial loophole is a policy change that is anticipated by firms, in markets where carriage

decisions of an RSN will change there will be potential equilibrium responses in all of these variables.

Computing counterfactual equilibria is currently ongoing.

In Table 5 we also report market outcomes (shares and surplus) in each market if exclusion

did occur: i.e., if the MVPD that owned each channel (reported below each channel in the table)

excluded its rival distributors in each market. Unsurprisingly, in all markets with exclusion, satellite

and consumer surplus is predicted to fall, as does satellite market share.

7 Concluding Remarks

This paper examined vertical integration of high value sports content in the US cable and satel-

lite television industry. Our framework accounts for consumer choice over downstream distribu-

tors, consumer viewership decisions over content, downstream pricing and carriage decisions, and

upstream-downstream bargaining over input fees. The framework allows for vertical integration to

31

Table 5: Removing Program Access Rules in 2007

Exclusion? Market Share SurplusλR = .79 λR = 1 w/o exc. exc. change w/o exc. exc. change

CSN Bay Area Yes Yes Int. Cable: 0.615 0.634 2.97% 8.202 8.461 3.16%Comcast Satellite 0.211 0.189 -10.45% 1.882 1.689 -10.26%

Pop. 5676023 Consumer: - - - 25.178 24.294 -3.51%

CSN CA No Yes Int. Cable: 0.605 0.609 0.67% 8.892 8.941 0.55%Comcast Satellite 0.212 0.207 -2.42% 1.899 1.853 -2.41%

Pop. 4623318 Consumer: - - - 24.391 24.221 -0.70%

CSN Chicago Yes Yes Int. Cable: 0.597 0.614 2.89% 6.861 7.103 3.52%Comcast Satellite 0.209 0.189 -9.36% 1.965 1.789 -8.98%

Pop. 5041614 Consumer: - - - 23.885 23.307 -2.42%

CSN Mid-Atl. No No Int. Cable: 0.662 0.674 1.68% 6.056 6.139 1.37%Comcast Satellite 0.165 0.152 -7.81% 1.644 1.519 -7.63%

Pop. 4423934 Consumer: - - - 25.378 25.020 -1.41%

CSN NE No Yes Int. Cable: 0.646 0.659 1.98% 9.040 9.205 1.83%Comcast Satellite 0.116 0.100 -13.10% 2.801 2.701 -3.59%

Pop. 4734329 Consumer: - - - 22.532 22.215 -1.41%

CSN NW Yes Yes Int. Cable: 0.598 0.632 5.70% 7.934 8.433 6.29%Comcast Satellite 0.254 0.217 -14.47% 2.206 1.889 -14.35%

Pop. 3275967 Consumer: - - - 36.672 35.731 -2.57%

CSS No No Int. Cable: 0.565 0.573 1.42% 6.330 6.384 0.85%Comcast, Charter Satellite 0.263 0.254 -3.61% 2.801 2.701 -3.59%

Pop. 10800000 Consumer: - - - 30.381 30.096 -0.94%

Cox Sports TV No No Int. Cable: 0.554 0.562 1.34% 4.060 4.064 0.10%Cox Satellite 0.208 0.200 -4.00% 1.977 1.902 -3.76%

Pop. 647210 Consumer: - - - 24.552 24.289 -1.07%

MSG No No Int. Cable: 0.687 0.697 1.45% 7.001 7.084 1.19%Cablevision Satellite 0.144 0.135 -6.11% 1.365 1.253 -8.23%

Pop. 11400000 RSN: - - - 0.598 0.607 1.56%

MSG Plus No Yes Int. Cable: 0.687 0.695 1.11% 6.816 6.916 1.47%Cablevision Satellite 0.144 0.135 -6.11% 1.365 1.283 -6.02%

Pop. 10800000 Consumer: - - - 27.959 27.739 -0.79%

Root NW No No Cable: 0.617 0.608 -1.40% 11.074 10.942 -1.19%DirecTV Int. Satellite 0.222 0.223 0.52% 1.066 1.085 1.79%

Pop. 3275967 Consumer: - - - 32.936 32.151 -2.38%

Root Pitt. No No Cable: 0.651 0.632 -2.84% 11.407 11.155 -2.21%DirecTV Int. Satellite 0.166 0.168 1.29% 1.367 1.380 0.98%

Pop. 8215501 Consumer: - - - 25.704 24.588 -4.34%

Root Rocky Mtn. No No Cable: 0.545 0.540 -0.91% 8.656 8.598 -0.68%DirecTV Int. Satellite 0.328 0.329 0.31% 1.736 1.752 0.90%

Pop. 4113810 Consumer: - - - 37.914 37.596 -0.84%

SportsNet NY No No Int. Cable: 0.687 0.693 0.86% 3.331 3.353 0.64%Comcast, TWC Satellite 0.144 0.137 -4.84% 1.365 1.301 -4.67%Pop. 11400000 Consumer: - - - 27.959 27.764 -0.70%

Notes: This table presents changes in the model’s predicted market shares and surplus under a counterfactual scenariowhere Int. RSN’s are permitted to refuse to deal with rivals. The MVPD listed under the channel name is the owner ofthe channel. The columns “exc.” and “w/o exc.” indicate scenarios with exclusion and without exclusion of satellite,and “change” reports the change. Surplus calculations for integrated MVPDs include surplus for the integrated RSN;all surplus figures are reported holding fixed exiting input fees and prices, and in units of $ per month per capita.

32

reduce double marginalization, to cause foreclosure of rivals to integrated content or to raise rivals’

costs of integrated content, and for the possibility that divisions within the integrated firm do not

perfectly internalize their actions on one another. We use the estimated model examine the effec-

tiveness of regulatory policy towards integrated sports content. We find that relaxing regulations

to allow exclusive dealing would result in foreclosure of cable integrated RSN’s to satellite providers

in a handful of large markets including Chicago and the San Francisco Bay Area. We predict such

foreclosure would decrease consumer surplus by roughly one to two percent per capita as consumers

with strong tastes for both satellite television and regional sports are unable to consume regional

sports on satellite.

This research can be extended in a number of directions. First, measuring the strength of this

set of effects in other industries would be important. Second, allowing for richer sets of behavior

on both the efficiency side and the strategic side could be important. For example, do vertically

integrated firms facilitate information sharing along the supply chain, and is this good or bad

for consumers? Third, incorporating dynamic effects of vertical integration such as changes in

investment incentives or post-merger entry would be an important step.

33

References

Asker, John. (2004), Measuring Advantages from Exclusive Dealing. Unpublished.

Berry, Steven, Levinsohn, James and Pakes, Ariel. (1995). ‘Automobile Prices in Market

Equilibrium’, Econometrica 63(4), 841–890.

Chen, D. and Waterman, D. (2007). ‘Vertical Ownership, Program Network Carriage, and

Tier Positioning in Cable Television: An Empirical Study’, Review of Industrial Organization

30(3), 227–251.

Chen, Yongmin. (2001). ‘On vertical mergers and their competitive effects’, RAND Journal of

Economics 32(4), 667–685.

Chipty, Tasneem. (2001). ‘Vertical Integration, Market Foreclosure, and Consumer Welfare in

the Cable Television Industry’, American Economic Review 91(3), 428–453.

Collard-Wexler, Allan, Gowrisankaran, Gautam and Lee, Robin S. (2014), Bargaining

in Bilateral Oligopoly: An Alternating Offers Representation of the “Nash-in-Nash” Solution.

NBER Working Paper 20641.

Conlon, Christopher T. and Mortimer, Julie H. (2013), All Units Discount: Experimental

Evidence from the Vending Industry. Unpublished.

Crawford, Gregory S., Lee, Robin S., Vieira, Breno, Whinston, Michael and Yu-

rukoglu, Ali. (2014), Channel 5 or 500? Vertical Integration, Favoritism, and Discrimination

in Multichannel Television. Unpublished.

Crawford, Gregory S. and Yurukoglu, Ali. (2012). ‘The Welfare Effects of Bundling in

Multichannel Television Markets’, American Economic Review 102(2), 643–685.

Draganska, Michaela, Klapper, Daniel and Villas-Boas, Sofia B. (2010). ‘A Larger Slice

or a Larger Pie? An Empirical Investigation of Bargaining Power in the Distribution Channel’,

Marketing Science 29(1), 57–74.

FCC. (2012), Report and Order in MB Docket Nos. 12-68, 07-18, 05-192; Further Notice of Pro-

posed Rulemaking in MB Docket No. 12-68; Order on Reconsideration in MB Docket No. 07-29,

Technical report. FCC 12-123, released October 5, 2012.

Gowrisankaran, Gautam, Nevo, Aviv and Town, Robert. (forthcoming). ‘Mergers When

Prices Are Negotiated: Evidence from the Hospital Industry’, American Economic Review .

Grennan, Matthew. (2013). ‘Price Discrimination and Bargaining: Empirical Evidence from

Medical Devices’, American Economic Review 103(1), 147–177.

34

Grossman, S. and Hart, O. (1986). ‘The Costs and Benefits of Ownership: A Theory of Vertical

and Lateral Integration’, Journal of Political Economy 94, 691–796.

Hart, Oliver and Tirole, Jean. (1990). ‘Vertical Integration and Market Foreclosure’, Brookings

Papers on Economic Activity. Microeconomics 1990, 205–286.

Hastings, J. and Gilbert, R. (2005). ‘Market Power, Vertical integration and the Wholesale

Price of Gasoline’, Journal of Industrial Economics 53(4), 469–492.

Ho, Kate and Lee, Robin S. (2013), Insurer Competition and Negotiated Hospital Prices. NBER

Working Paper 19401.

Horn, Hendrick and Wolinsky, Asher. (1988). ‘Bilateral Monopoly and Incentives for Merger’,

The RAND Journal of Economics 19, 408–419.

Hortacsu, Ali and Syverson, Chad. (2007). ‘Cementing Relationships: Vertical Integration,

Foreclosure, Productivity, and Prices’, Journal of Political Economy 115(2), 250–301.

Houde, Jean-Francois. (2012). ‘Spatial Differentiation and Vertical Mergers in Retail Markets

for Gasoline’, American Economic Review 102(5), 2147–2182.

Krattenmaker, Thomas G. and Salop, Steven C. (1986). ‘Anticompetitive Exclusion: Raising

Rivals’ Costs to Achieve Power over Price’, The Yale Law Journal 2, 209–293.

Lee, Robin S. (2013). ‘Vertical Integration and Exclusivity in Platform and Two-Sided Markets’,

American Economic Review 103(7), 2960–3000.

McAfee, R. Preston and Schwartz, Marius. (1994). ‘Opportunism in Multilateral Verti-

cal Contracting: Nondiscrimination, Exclusivity, and Uniformity’, American Economic Review

84(1), 210–230.

Nocke, Volker and White, Lucy. (2007). ‘Do Vertical Mergers Facilitate Upstream Collusion’,

American Economic Review 97(4), 1321–1339.

Ordover, J., Saloner, G. and Salop, S. (1990). ‘Equilibrium Vertical Foreclosure’, American

Economic Review 80, 127–142.

Pakes, Ariel, Porter, Jack, Ho, Kate and Ishii, Joy. (forthcoming). ‘Moment Inequalities

and Their Application’, Econometrica .

Perry, Martin K. (1990), Vertical Integration: Determinants and Effects, in Richard

Schmalensee and Robert Willig., eds, ‘Handbook of Industrial Organization’, Vol. 1, North

Holland, Amsterdam.

Riordan, Michael H. (2008), Competitive Effects of Vertical Integration, in Paolo Buccirossi.,

ed., ‘Handbook of Antitrust Economics’, MIT Press, Cambridge, MA.

35

Salop, Steven C. and Scheffman, David T. (1983). ‘Raising Rivals’ Costs’, American Economic

Review 73(2), 267–271.

Spengler, Joseph J. (1950). ‘Vertical integration and antitrust policy’, The Journal of Political

Economy pp. 347–352.

Suzuki, Ayako. (2009). ‘Market foreclosure and vertical merger: A case study of the vertical

merger between Turner Broadcasting and Time Warner’, International Journal of Industrial

Organization 27(4), 523–543.

Villas-Boas, Sofia B. (2007). ‘Vertical Relationships between Manufacturers and Retailers: In-

ference with Limited Data’, Review of Economic Studies 74(2), 625–652.

Waterman, D. H. and Weiss, A. A. (1996). ‘The Effects of Vertical Integration between Cable

Television Systems and Pay Cable Networks’, Journal of Econometrics 72(1-2), 357–95.

Willamson, Oliver. (1985), The economic institutions of capitalism, New York, NY.

36

A Further Estimation and Computational Details

A.1 Solving for Negotiated Input Fees and Bundle Marginal Costs

We will omit the subscript on Ψfct for the expressions in this subsection.Consider MVPD f bargaining with channel c over input fee τfct. Closed form expressions for MVPD

and channel “GFT” terms defined in (8) can be derived as follows:

GFTMfct =

∑m∈Mfct

[[µfctDfmt −D\fcfmt

]τfct + µfct(Dfmt +

∑g 6=f :c∈Bgmt

[∆fcDgmt])acmt (16)

+ µfct

∑g 6=f :c∈Bgmt

[∆fcDgmt]τgct +∑

d∈Vft\c

∑g∈Fmt:d∈Bgmt

[∆fcDgmt]µfdt(τgdt + admt)

+ [∆fcDfmt](pfmt −mcfmt

)]

GFTCfct =

∑m∈Mfct

[(Dfmt − µfctD

\fcfmt)τfct + (Dfmt +

∑g 6=f :c∈Bgmt

[∆fcDgmt])acmt +∑

g 6=f :c∈Bgmt

[∆fcDgmt](τgct)

(17)

+∑

g∈Fmt

λR:fct[∆fcDgmt]∑

d∈Bgmt\c

µCcdt(τgdt + admt) +

∑g∈Fmt

µgctλR:fct[∆fcDgmt](pgmt −mcgmt

)]

where: D\fcfmt is the demand for f in market m if it dropped channel c; λR:fct = λR if f and c are not

integrated, and λR:fct = 1 otherwise; µfct = µ× ofct; and µCcdt = µ× oC

cdt.Assume now that c is either non-integrated or integrated with either f or f ′. Using (16) and (17), the

Nash Bargaining FOC given by (10) (GFTCfct = ΨGFTM

fct) for f and c bargaining can be expressed as:

τfct∑

m∈Mfct

[Dfmt(1−Ψµfct) +D

\fcfmt(Ψ− µfct)

]+

∑g 6=f :c∈Bgmt

τgct∑

m∈Mfct

(1−Ψµfct)[∆fcDgmt] (18)

+∑

g∈Fmt

∑d∈Bgmt\c

τgdt(µCcdtλR:fct −Ψµfdt)

∑m∈Mfct

[∆fcDgmt]

+(Ψ− µfct)∑

m∈Mfct

mcfmt[∆fcDfmt]− µf ′ctλR∑

m∈Mfct

mcf ′mt[∆fcDf ′mt] =

∑m∈Mfct

[(Ψ− µfct)[∆fcDfmt]pfmt − µf ′ctλR[∆fcDf ′mt]pf ′mt

]

−∑

m∈Mfct

[acmt

((1−Ψµfct)Dfmt + (1−Ψµfct)

∑g 6=f :c∈Bgmt

[∆fcDgmt])

+∑

g∈Fmt

∑d∈Bgmt\c

admt(µCcdtλR:fct −Ψµfdt)([∆fcDgmt])

]

which will hold for all MVPD-channel (f, c) pairs.We can also re-express the optimal prices set by each MVPD given by the pricing FOC in (5) as:[ ∑g∈Fmt

∂sgmt

∂pfmt

∑c∈Bgmt

µfctτgct

]− ∂sfmt

∂pfmtmcfmt = −

[sfmt +

∂sfmt

∂pfmtpfmt +

∑g∈Fmt

∂sgmt

∂pfmt

∑c∈Bgmt

µfctacmt

](19)

37

which will hold for all markets and firms active (∀m, f ∈ Fmt).Using (18) and (19), which have expressed input fees and marginal costs on the LHS as a function of

demand parameters, prices, and advertising rates, the vector of input fees and bundle marginal costs can besolved explicitly via matrix inversion.

Implementation. We first solve for {τfct}∀f,t,c∈CRSNt

for all RSNs and for {mcfmt}∀fmt using (18) and(19). Once these have been recovered, we use our estimates to recover {τfct}∀ft,c/∈CRSN

tfor non-RSN channels,

which we have assumed to be non-integrated (or not internalize integrated unit profits), via matrix inversionon the following:

τfct∑

m∈Mfct

[Dfmt + ΨD

\fcfmt

]+

∑g 6=f :c∈Bgmt

τgct∑

m∈Mfct

[∆fcDgmt] = (20)

∑m∈Mfct

[(Ψ)[∆fcDfmt](pfmt −mcfmt)

]+∑

g∈Fmt

∑d∈Bgmt\c

µfdtΨτ̂gdt∑

m∈Mfct

[∆fcDgmt]

−∑

m∈Mfct

[acmt

(Dfmt +

∑g 6=f :c∈Bgmt

[∆fcDgmt])

+∑

g∈Fmt

∑d∈Bgmt\c

admt(−Ψµfdt)([∆fcDgmt])

]

At this stage, since we are focused on recovering estimated input costs for non-RSN channels c, it will bethe case that µfct = 0 ∀ft, c /∈ CRSN

t . Also, the only input costs that enter into the calculation are for RSNson the RHS of (20)); thus, when specifying f and c’s bargain, we use estimates of these RSN input costsrecovered in the first stage to account for changes in realized total input costs from f ’s other integratedchannels.

A.2 Computation of Disagreement Payoffs

Computation of several moments requires estimating ∆fc[ΠMfmt(Bmt,pmt, {τ̂fct, τ−fc,t})] and

∆fc[ΠCcmt(Bmt,pmt, {τ̂fct, τ−fc,t};λR)] for each MVPD f and channel c that contract in each period. These

“gains from trade” for each pair are comprised of agreement and disagreement profits.Profits from agreement (as a function of θ) can be computed from observed prices and bundle composition

using MVPD and Channel profits specified by (4) and (7). Profits from disagreement between MVPD f andchannel c are recomputed in each market given the following assumptions:

1. Bundle composition does not change for other MVPDs: B′gmt = Bgmt ∀ g 6= f ; bundles for MVPD fjust drop c, but do not adjust otherwise;

2. Input prices τ̂−fc,t for all other MVPD-conglomerate pairs do not adjust;

3. Bundle prices for satellite and cable providers do not adjust.

The second and third assumptions are consistent with the timing of our game and the simultaneous deter-mination of input and bundle prices.

A.3 Recomputing Counterfactual Equilibria when Channels are Added or Re-moved from Satellite

When we explore counterfactuals when a RSN channel c is either added or removed from satellite providers(and potentially un-integrated), we compute market outcomes when input and bundle prices are allowed tore-equilibrate. Note that this is different than in the previous subsection, where we explore the computationof disagreement points which occur off the equilibrium path, since here changes are anticipated by all players(e.g., if the terrestrial loophole were closed). We assume that:

1. satellite distributors either carry or do not carry c in all (relevant) markets, and (with national pricing)do not change the prices of its bundles;

38

2. cable systems may change their prices (since demand elasticities may be affected by changes in carriage)but do not change any carriage or bundling decisions;

3. input prices of RSNs (but not national channels) are allowed to adjust.

We compute the new counterfactual equilibrium where c is either now supplied or removed from satellitein a given period t as follows:

1. We recover estimates of non-RSN costs for each MVPD in the relevant markets:

m̂c\Rfmt ≡ m̂cfmt −

∑c∈BR

fmt

τ̂fct ∀m ∈Mc, f ∈ Fmt (21)

where BRfmt is the observed set of RSNs carried by f in market m.

2. Given new bundles {BR,CFfmt } and potentially new values for {λCF

R:fct, µCFfct }, we iterate on the follow-

ing until we obtain convergence on counterfactual input prices {τCFfct }, bundle prices {pCF

fmt}, bundle

demands {DCFfmt}, and elasticities {∂sCF

fmt/∂pgmt}:

(a) Solve for the values of {τCFfct }c∈CRSN

tgiven values of {DCF

fmt}, {∂sCFfmt/∂pgmt}, {m̂c\Rfmt}, µ, λR,

Ψ using the following system of equations:

τCFfct

∑m∈Mfct

[(1 + Ψ)(1− µfct)D

CFfmt

]+

∑g 6=f :c∈BR,CF

gmt

τCFgct

∑m∈Mfct

(1−Ψµfct − µgctλR)[∆fcDCFgmt] (22)

+∑

g∈Fmt

∑d∈BR,CF

gmt \c

τCFgdt (µC

cdtλR:fct −Ψµfdt − µgctλR)∑

m∈Mfct

[∆fcDCFgmt] =

∑m∈Mfct

[(Ψ− µfct)(p

CFfmt − m̂c

\Rfmt)[∆fcD

CFfmt]− µf ′ctλR(pCF

fmt − m̂c\Rf ′mt)[∆fcD

CFf ′mt]

]

−∑

m∈Mfct

[acmt

((1−Ψµfct)D

CFfmt + (1−Ψµfct)

∑g 6=f :c∈BR,CF

gmt

[∆fcDCFgmt]

)

+∑

g∈Fmt

∑d∈BR,CF

gmt \c

admt(µCcdtλR:fct −Ψµfdt)([∆fcD

CFgmt])

]∀f, c

where, again, f and f ′ represent the MVPDs with which c is potentially integrated.

(b) Market by market, update bundle prices {pCFfmt} for all cable distributors to maximize profits

given new values of {τCFfct }. Update bundle demands {DCF

fmt} and elasticities {∂sCFfmt/∂pgmt} at

the new computed prices.

39

B Additional Figures and Tables

Table 6: Regional Sports Networks Availability, Affiliate Fees, and Viewership

Kagan Kagan NielsenAvailability Affiliate Fees Viewing

Systems HH All Has HasServed Served Years Mean StDev Min Max Obs HH DTV Dish

Comcast RSNsComcast SportsNet Bay Area 137 4.7 11 $1.70 $0.53 $1.01 $2.52 720 0.41 0.45 0.33Comcast SportsNet California 1,960 59.4 7 $0.91 $0.14 $0.75 $1.10 720 0.17 0.17 0.17Comcast SportsNet Chicago 67 0.9 7 $2.02 $0.18 $1.90 $2.37 360 0.54 0.59 0.36Comcast SportsNet Mid-Atlantic 23 1.7 11 $2.03 $0.74 $0.85 $3.10 1,440 0.13 0.09 0.03Comcast SportsNet New England 15 1.0 11 $1.26 $0.32 $0.90 $1.89 1,080 0.27 0.30 0.17Comcast SportsNet Northwest 137 4.7 4 $1.93 $0.09 $1.81 $2.04 — — — —Comcast SportsNet Philadelphia 135 10.0 11 $1.94 $0.61 $1.05 $2.85 360 0.91 0.06 0.05Comcast SportsNet Southwest 335 5.7 — — — — — — — — —Comcast/Charter Sports Southeast 194 6.2 11 $0.36 $0.09 $0.20 $0.50 3,600 0.04 0.00 0.00The mtn 195 7.0 5 $0.20 $0.02 $0.19 $0.23 720 0.04 0.05 0.00

News Corp RSNsFox Sports Arizona 106 3.7 11 $1.58 $0.50 $0.82 $2.28 — — — —Fox Sports Chicago 342 4.8 7 $1.45 $0.44 $1.08 $2.13 — — — —Fox Sports Detroit 284 5.3 11 $1.75 $0.45 $1.05 $2.34 360 1.02 0.94 0.68Fox Sports Florida 152 6.7 11 $1.34 $0.33 $0.90 $1.95 2,160 0.14 0.12 0.12Fox Sports Houston 48 3.3 — — — — — — — — —Fox Sports Midwest 695 7.4 11 $1.42 $0.44 $0.57 $2.01 1,800 0.31 0.31 0.26Fox Sports North 620 4.5 11 $1.97 $0.60 $1.15 $2.88 720 0.79 1.04 0.70Fox Sports Ohio 306 7.0 11 $1.61 $0.49 $0.75 $2.42 2,160 0.34 0.31 0.29Fox Sports South 905 15.3 17 $1.63 $0.52 $0.52 $2.17 3,600 0.13 0.08 0.07Fox Sports Southwest 924 12.7 11 $1.68 $0.50 $0.80 $2.43 5,040 0.14 0.15 0.12Fox Sports West 167 9.2 11 $1.80 $0.44 $0.87 $2.35 1,080 0.16 0.12 0.07Fox Sports Wisconsin 136 2.2 — — — — — — — — —Big Ten Network 1,960 59.4 — — — — — — — — —Prime Ticket (New) 132 8.2 11 $1.52 $0.46 $0.60 $2.07 720 0.16 0.12 0.09SportSouth (New) 532 11.3 11 $0.31 $0.13 $0.15 $0.52 — — — —Sun Sports 234 8.3 11 $1.36 $0.54 $0.55 $2.27 2,160 0.20 0.16 0.12

Liberty RSNsRoot Sports Northwest 281 5.4 11 $1.73 $0.52 $0.70 $2.54 — — — —Root Sports Pittsburgh 316 4.5 11 $1.81 $0.53 $1.05 $2.55 — — — —Root Sports Rocky Mountain 479 5.4 11 $1.58 $0.42 $0.75 $2.06 — — — —

Cablevision RSNsMadison Sq. Garden (MSG) 219 9.9 11 $1.82 $0.30 $1.45 $2.44 1,080 0.23 0.24 0.17MSG Plus 165 7.5 11 $1.24 $0.15 $1.01 $1.61 360 0.07 0.05 0.06

Cox RSNsChannel 4 San Diego 15 1.0 11 $0.87 $0.26 $0.53 $1.32 360 0.48 0.03 0.00Cox Sports Television 70 2.1 9 $0.55 $0.05 $0.50 $0.64 360 0.22 0.01 0.08

Time Warner RSNsMetro Sports Network 8 0.6 — — — — — — — — —SportsNet New York 314 20.1 5 $1.91 $0.18 $1.71 $2.20 1,080 0.13 0.13 0.09

Independent/Other RSNsAltitude Sports & Entertainment 130 2.8 7 $1.99 $0.29 $1.70 $2.47 360 0.24 0.21 0.22Bright House Sports Network — — — — — — — 360 0.02 0.00 0.00Empire Sports Network 87 1.9 — — — — — — — — —Mid-Atlantic Sports Network (MASN) 109 5.2 6 $1.58 $0.12 $1.45 $1.77 1,440 0.13 0.10 0.13New England Sports Network (NESN) 213 4.5 11 $1.99 $0.49 $1.30 $2.72 1,080 0.95 1.00 0.48Royals Sports 18 0.2 6 $0.19 $0.02 $0.16 $0.21 — — — —SportsTime Ohio 196 9.0 5 $1.51 $0.17 $1.30 $1.73 720 0.33 0.40 0.20Yankees Entertainment & Sports (YES) 304 15.8 9 $2.13 $0.41 $1.18 $2.62 1,440 0.27 0.30 0.00

Notes: Reported are availability, affiliate fees and average viewing of the major Regional Sports Networks (RSNs)in the United States. Affiliate fees are the monthly per-subscriber fees paid by cable and satellite distributors totelevision networks for the right to distribute the network’s programming to subscribers. Availability and affiliatefee information is provided by SNL Kagan as part of its Media & Communications Package. RSN viewership is fromNielsen and covers 2000-2010.

40

Fig

ure

8:R

SN

Ow

ner

ship

Notes:

Rep

ort

edare

the

ver

tica

low

ner

ship

stakes

hel

dby

majo

rdis

trib

uto

rsof

cable

and

sate

llit

ete

levis

ion

serv

ice

inR

egio

nal

Sp

ort

sN

etw

ork

s(R

SN

s).

This

data

was

collec

ted

by

hand

from

com

pany

stock

filings

and

indust

ryso

urc

es.

The

owner

ship

share

for

each

dis

trib

uto

ris

rep

ort

edand

indiv

idual

owner

s(o

rco

mbin

ati

ons

of

owner

s)are

colo

r-co

ded

acc

ord

ing

toth

ele

gen

d.

Gra

ysh

adin

gco

rres

ponds

toa

yea

rin

whic

hth

egiv

enR

SN

has

not

yet

ente

red

or

has

exit

edth

em

ark

et.

Whit

eb

oxes

corr

esp

ond

toyea

rsof

act

ive

op

erati

on

for

an

RSN

wit

hout

aver

tica

low

ner

ship

affi

liati

on.

41

Table 7: Sample Statistics - Prices, Market Shares, and Channels

Weighted byUnweighted Households

Obs Mean StdDev Min Max Mean StdDev Min Max

Total Markets 6,138 6,138Average Households (M) 6,138 31.5

CableYear 6,138 2004 2.9 2000 2010 2004 2.8 2000 2010Price 6,138 $51.40 $10.33 $8.67 $130.96 $53.02 $8.84 $8.67 $130.96Market Share 6,138 0.624 0.161 0.005 0.965 0.630 0.137 0.005 0.965Cable Networks 6,138 42.6 15.4 0 87 44.9 14.0 0 87RSNs 6,138 1.6 0.9 0 5 1.8 0.9 0 5Total Channels 6,138 44.2 15.9 1 90 46.6 14.5 1 90

DirecTVYear 6,138 2004 2.9 2000 2010 2004 2.8 2000 2010Price 6,138 $53.25 $6.57 $46.05 $76.73 $53.27 $6.34 $46.05 $76.73Market Share 6,138 0.092 0.062 0.002 0.499 0.094 0.064 0.002 0.499Cable Networks 6,138 80.5 10.3 66 97 81.2 10.1 66 97RSNs 6,138 1.7 0.9 0 6 1.9 0.9 0 6Total Channels 6,138 82.2 10.5 66 103 83.0 10.3 66 103

DishYear 6,138 2004 2.9 2000 2010 2004 2.8 2000 2010Price 6,138 $53.89 $4.75 $44.28 $68.33 $53.96 $4.53 $44.28 $68.33Market Share 6,138 0.064 0.055 0.000 0.406 0.059 0.052 0.000 0.406Cable Networks 6,138 70.8 13.2 54 91 71.8 12.9 54 91RSNs 6,138 1.6 0.8 0 5 1.7 0.7 0 5Total Channels 6,138 72.4 13.3 54 96 73.5 13.0 54 96

Notes: Reported are the price, market share, and cable, Regional Sport Network (RSN), and total channels for eachof the local cable operators and two national satellite providers serving each of our markets. Markets are definedas the set of continuous zip codes within a cable system facing the same portfolio of competitors. We exclude (therelatively few) markets facing competition between cable operators. All the data cover the years 2000-2010. To beincluded, we required information on each of price, market share, and channels. Cable system subscriber and channelinformation is from the Nielsen FOCUS dataset. Cable system price information is drawn from the Internet Archive,newspaper reports, and the TNS Bill Harvesting database. Satellite system channel and price information is drawnfrom the Internet Archive. Cable and satellite subscriber market shares are estimated from the MRI (2000-2007) andSimmons (2008-2010) household surveys. We restrict attention to those markets with at least 5 observations in anyyear. See the text for more details.

42

Table 8: Sample Statistics - National Cable Channel Affiliate Fees and Viewership, Part 1

Affiliate Fees ViewershipKagan Nielsen Combined MRI / Simmons

PercentYears Mean StDev Min Max Obs Mean Obs Mean SDev Positive

Channels (A-L)ABC Family Channel 11 $0.19 $0.02 $0.16 $0.22 747 0.418 277,535 0.344 1.149 0.176AMC 11 $0.22 $0.02 $0.20 $0.25 747 0.491 277,535 0.351 1.183 0.156Animal Planet 11 $0.07 $0.01 $0.06 $0.09 747 0.275 277,535 0.344 1.108 0.203A&E 11 $0.21 $0.03 $0.16 $0.26 747 0.664 277,535 0.472 1.373 0.230BBC America 11 $0.09 $0.03 $0.03 $0.12 703 0.053 225,618 0.091 0.617 0.041BET 11 $0.14 $0.02 $0.11 $0.17 747 0.382 277,535 0.184 1.017 0.070Bio 11 $0.07 $0.03 $0.00 $0.11 447 0.082 98,567 0.104 0.618 0.023Bloomberg Television 11 $0.04 $0.02 $0.02 $0.06 — — 150,165 0.029 0.373 0.010Boomerang 10 $0.05 $0.03 $0.00 $0.08 280 0.131 — — — —Bravo 11 $0.15 $0.03 $0.11 $0.20 747 0.277 277,535 0.169 0.804 0.092Cartoon Network 11 $0.14 $0.03 $0.08 $0.18 747 0.989 277,535 0.231 1.098 0.106CMT 11 $0.06 $0.02 $0.01 $0.08 — — 277,535 0.120 0.732 0.067CNBC 11 $0.24 $0.04 $0.16 $0.30 747 0.217 277,535 0.313 1.185 0.170CNN 11 $0.43 $0.05 $0.35 $0.52 747 0.550 277,535 0.701 1.744 0.319CNN en Espanol — — — — — 463 0.013 — — — —CNN International 11 $0.11 $0.02 $0.09 $0.13 567 0.012 — — — —Comedy Central 11 $0.11 $0.02 $0.08 $0.14 747 0.449 277,535 0.280 0.997 0.162Discovery Channel 11 $0.27 $0.04 $0.22 $0.35 747 0.535 277,535 0.628 1.462 0.327Disney Channel 11 $0.81 $0.06 $0.75 $0.91 747 1.171 277,535 0.246 1.074 0.116E! Entertainment TV 11 $0.19 $0.02 $0.15 $0.21 747 0.315 277,535 0.201 0.788 0.137ESPN 11 $2.81 $1.12 $1.14 $4.34 747 0.836 277,535 0.675 1.767 0.257ESPN 2 11 $0.37 $0.14 $0.17 $0.58 747 0.262 277,535 0.334 1.220 0.151ESPN Classic Sports 11 $0.14 $0.03 $0.10 $0.18 636 0.037 277,535 0.072 0.521 0.047ESPN deportes — — — — — 280 0.035 — — — —ESPNews 11 $0.10 $0.06 $0.02 $0.17 636 0.043 277,535 0.143 0.782 0.084ESPNU 6 $0.14 $0.03 $0.10 $0.17 280 0.037 — — — —Fine Living Network — — — — — 55 0.003 150,165 0.025 0.324 0.009FitTV 11 $0.05 $0.02 $0.02 $0.07 205 0.005 — — — —Flix — — — — — — — 101,275 0.013 0.165 0.004Food Network 11 $0.06 $0.03 $0.03 $0.14 747 0.411 277,535 0.396 1.364 0.175Fox News Channel 11 $0.32 $0.18 $0.17 $0.70 747 0.785 277,535 0.697 1.961 0.267Fuse 11 $0.06 $0.01 $0.05 $0.08 747 0.024 225,618 0.018 0.308 0.009FX 11 $0.34 $0.06 $0.27 $0.43 747 0.463 277,535 0.258 0.976 0.137G4 9 $0.07 $0.02 $0.05 $0.09 591 0.051 225,618 0.036 0.411 0.016GSN 11 $0.07 $0.03 $0.04 $0.10 747 0.154 277,535 0.088 0.703 0.036Golf Channel 11 $0.20 $0.05 $0.13 $0.26 580 0.065 277,535 0.084 0.633 0.041Hallmark Channel 11 $0.04 $0.02 $0.01 $0.06 699 0.307 225,618 0.301 1.268 0.088Headline News — — — — — 747 0.214 277,535 0.278 0.983 0.173HGTV 11 $0.08 $0.04 $0.03 $0.14 747 0.500 277,535 0.397 1.446 0.162History Channel 11 $0.18 $0.04 $0.13 $0.23 747 0.531 277,535 0.531 1.462 0.251HSN — — — — — 580 0.038 252,217 0.044 0.395 0.031IFC 11 $0.18 $0.01 $0.17 $0.19 — — 277,535 0.045 0.424 0.023Investigation Discovery 11 $0.04 $0.03 $0.00 $0.07 441 0.121 174,621 0.067 0.628 0.018Lifetime 11 $0.21 $0.06 $0.13 $0.29 747 0.679 277,535 0.554 1.650 0.199Lifetime Movie Network 11 $0.07 $0.03 $0.00 $0.09 328 0.185 225,618 0.250 1.174 0.068

Notes: Reported are affiliate fees and average viewing of the major cable television networks included in our demandsystem. Affiliate fees are the monthly per-subscriber fees paid by cable and satellite distributors to television networksfor the right to distribute the network’s programming to subscribers. Affiliate fee information is provided by SNLKagan as part of its Media & Communications Package. Nielsen viewership data reports the average rating oneach channel across between 44 and 56 Designated Market Areas (DMAs) between 2000 and 2010. MRI / Simmonsviewership data reports the average viewership and the percent of households with any (positive) viewership of eachchannel by households in the MRI (2000-2007) and Simmons (2008-2010) household surveys.

43

Table 9: Sample Statistics - National Cable Channel Affiliate Fees and Viewership, Part 2

Affiliate Fees ViewershipKagan Nielsen Combined MRI / Simmons

PercentYears Mean StDev Min Max Obs Mean Obs Mean SDev Positive

Channels (M-Z)MSNBC 11 $0.14 $0.02 $0.12 $0.17 747 0.343 277,535 0.330 1.181 0.182MTV 11 $0.27 $0.05 $0.20 $0.35 747 0.568 277,535 0.235 0.983 0.127MTV Hits 11 $0.01 $0.00 $0.01 $0.01 280 0.030 — — — —MTV Jams 11 $0.01 $0.01 $0.01 $0.02 280 0.038 — — — —MTV2 11 $0.03 $0.01 $0.01 $0.05 601 0.082 277,535 0.070 0.542 0.042Nat Geo Wild 6 $0.07 $0.02 $0.04 $0.09 112 0.068 — — — —Nat Geo Channel 11 $0.17 $0.06 $0.00 $0.21 608 0.136 225,618 0.212 0.883 0.096NBA TV 11 $0.31 $0.06 $0.19 $0.37 280 0.035 — — — —NBC Sports / Versus 8 $0.50 $0.33 $0.11 $0.85 55 0.047 — — — —NFL Network 4 $0.45 $0.09 $0.32 $0.53 56 0.027 — — — —NHL Network 11 $0.11 $0.06 $0.00 $0.18 376 0.082 — — — —Nickelodeon 11 $0.37 $0.05 $0.29 $0.47 747 1.555 277,535 0.200 0.991 0.096NickToons TV 9 $0.05 $0.03 $0.00 $0.07 447 0.128 — — — —OWN 11 $0.06 $0.03 $0.00 $0.09 280 0.130 — — — —Outdoor Channel 11 $0.04 $0.01 $0.03 $0.05 — — 174,621 0.068 0.594 0.021Ovation 11 $0.06 $0.02 $0.03 $0.08 280 0.027 — — — —Oxygen 11 $0.07 $0.04 $0.00 $0.10 656 0.131 225,618 0.114 0.658 0.052ReelzChannel — — — — — 280 0.033 — — — —Science Channel 11 $0.04 $0.02 $0.00 $0.07 592 0.072 174,621 0.092 0.635 0.030ShopNBC — — — — — 280 0.025 — — — —SoapNet 11 $0.11 $0.05 $0.02 $0.15 656 0.135 174,621 0.109 0.833 0.022Speed Channel 11 $0.17 $0.03 $0.11 $0.21 747 0.091 277,535 0.097 0.679 0.046Style Network 11 $0.10 $0.04 $0.03 $0.14 646 0.063 225,618 0.040 0.416 0.019Sundance Channel 11 $0.23 $0.04 $0.16 $0.27 — — 174,621 0.037 0.397 0.012SyFy 11 $0.17 $0.04 $0.12 $0.22 747 0.427 277,535 0.301 1.207 0.126TBS 11 $0.37 $0.12 $0.19 $0.54 747 0.905 277,535 0.497 1.345 0.243TechTV 4 $0.02 $0.01 $0.00 $0.03 47 0.006 51,917 0.012 0.202 0.002The Hub 11 $0.04 $0.02 $0.01 $0.06 441 0.037 — — — —TLC 11 $0.16 $0.01 $0.14 $0.17 747 0.422 277,535 0.342 1.151 0.173Toon Disney — — — — — 376 0.146 177,590 0.096 0.644 0.034Travel Channel 11 $0.07 $0.02 $0.04 $0.11 747 0.166 277,535 0.157 0.712 0.106truTV 11 $0.09 $0.01 $0.08 $0.10 747 0.384 277,535 0.233 1.081 0.101Turner Classic Movies 11 $0.22 $0.03 $0.16 $0.27 580 0.286 277,535 0.268 1.142 0.105TNT 11 $0.83 $0.16 $0.55 $1.10 747 1.219 277,535 0.592 1.553 0.263TV Guide Network 11 $0.03 $0.01 $0.02 $0.05 656 0.101 277,535 0.082 0.488 0.082TV Land 11 $0.08 $0.03 $0.01 $0.12 376 0.412 277,535 0.190 0.979 0.086TV One 7 $0.03 $0.03 $0.00 $0.08 280 0.129 123,885 0.050 0.572 0.008USA 11 $0.46 $0.07 $0.36 $0.57 747 1.081 277,535 0.503 1.442 0.230VH1 11 $0.12 $0.02 $0.09 $0.16 747 0.336 277,535 0.151 0.717 0.101VH1 Classic 11 $0.05 $0.01 $0.02 $0.07 55 0.024 149,303 0.044 0.422 0.016Weather Channel 11 $0.10 $0.01 $0.08 $0.12 747 0.234 204,189 0.380 0.879 0.266WE 11 $0.09 $0.01 $0.07 $0.11 328 0.084 225,618 0.096 0.621 0.041

Notes: Reported are affiliate fees and average viewing of the major cable television networks included in our demandsystem. Affiliate fees are the monthly per-subscriber fees paid by cable and satellite distributors to television networksfor the right to distribute the network’s programming to subscribers. Affiliate fee information is provided by SNLKagan as part of its Media & Communications Package. Nielsen viewership data reports the average rating oneach channel across between 44 and 56 Designated Market Areas (DMAs) between 2000 and 2010. MRI / Simmonsviewership data reports the average viewership and the percent of households with any (positive) viewership of eachchannel by households in the MRI (2000-2007) and Simmons (2008-2010) household surveys.

44

Table 10: Monthly WTP for Non-RSNs

Channel Name Mean WTP Fraction Positive Mean Among Positive

ABC Family Channel 0.744 0.699 1.065American Movie Classics AMC 0.770 0.627 1.228

Animal Planet 0.648 0.794 0.816Arts Entertainment AE 0.846 0.791 1.069

BET 0.490 0.504 0.971Bravo 0.360 0.504 0.715

Cartoon Network 1.212 0.598 2.027CMT 0.226 0.500 0.452

CNBC 0.457 0.663 0.689CNN 1.258 0.904 1.392

Comedy Central 0.692 0.704 0.983Discovery Channel 1.119 0.852 1.313

Disney Channel 0.675 0.583 1.158E Entertainment TV 0.407 0.545 0.747

ESPN 8.040 0.761 10.562ESPN 2 3.283 0.595 5.521

ESPN Classic 2.113 0.670 3.152Food Network 0.784 0.746 1.051

Fox News Channel 1.351 0.890 1.519FX 0.639 0.641 0.997

Golf Channel 0.278 0.397 0.702Hallmark Channel 0.694 0.595 1.166

Headline News 0.537 0.667 0.806HGTV 0.853 0.703 1.213

History Channel 0.989 0.866 1.142Lifetime 1.098 0.713 1.539MSNBC 0.612 0.703 0.870

MTV 0.665 0.605 1.098Nickelodeon 1.173 0.602 1.950SyFy, Sci-Fi 0.718 0.755 0.950

TBS 1.226 0.828 1.480TLC, The Learning Channel 0.603 0.692 0.872

truTV, Court TV 0.506 0.474 1.069Turner Classic Movies 0.761 0.805 0.946

Turner Network TV TNT 1.458 0.800 1.824USA 1.290 0.709 1.820VH1 0.488 0.551 0.885

Weather Channel 0.640 0.945 0.677

Notes: This table presented estimated mean monthly willingness-to-pay in dollars for the non-RSN’s. The firstcolumn is the unconditional mean. The second column is the fraction of consumers with positive valuations. Thethird column is the mean amongst those with positive valuations.

45

Tab

le11

:M

onth

lyW

TP

for

RS

Ns

RSN

Nam

eM

ean

WT

PF

ract

ion

Posi

tive

Mea

nA

mong

Posi

tive

Alt

itude

Sp

ort

sE

nte

rtain

men

t1.1

64

0.5

93

1.9

62

Channel

4San

Die

go,

4SD

3.6

85

0.9

03

4.0

80

CSN

Bay

Are

a,

Fox

Sp

ort

sNet

Bay

Are

a,

Sp

ort

sChannel

Paci

fic

2.6

90

0.8

84

3.0

42

CSN

Califo

rnia

1.0

42

0.6

01

1.7

35

CSN

Chic

ago

2.7

05

0.5

96

4.5

36

CSN

Mid

-Atl

anti

c,H

om

eT

eam

Sp

ort

s1.0

94

0.5

88

1.8

61

CSN

New

Engla

nd,

Fox

Sp

ort

sN

etN

ewE

ngla

nd,

Sp

ort

sChannel

New

Engla

nd,

Pri

smN

ewE

ngla

nd

2.5

41

0.6

02

4.2

22

CSN

Nort

hw

est

3.0

95

0.5

91

5.2

33

CSN

Philadel

phia

,Sp

ort

sChannel

Philadel

phia

,P

RIS

M7.0

79

0.8

00

8.8

51

Com

cast

Chart

erSp

ort

sSouth

east

CSS

1.0

05

0.6

02

1.6

71

Cox

Sp

ort

sT

elev

isio

n2.6

23

0.5

05

5.1

98

Fox

Sp

ort

sD

etro

it,

Pro

-Am

Sp

ort

sSyst

emP

ASS

2.4

98

0.6

06

4.1

23

Fox

Sp

ort

sF

lori

da,

Sp

ort

sChannel

Flo

rida

1.6

69

0.6

85

2.4

35

Fox

Sp

ort

sM

idw

est,

Pri

me

Sp

ort

sM

idw

est

1.1

94

0.7

04

1.6

97

Fox

Sp

ort

sN

ort

h,

Fox

Sp

ort

sM

inn

Wis

c,M

idw

est

Sp

ort

sC

hannel

1.3

47

0.6

95

1.9

39

Fox

Sp

ort

sO

hio

,Sp

ort

sChannel

Ohio

4.4

33

0.8

13

5.4

51

Fox

Sp

ort

sSouth

,Sp

ort

South

Old

0.4

13

0.5

99

0.6

91

Fox

Sp

ort

sSouth

wes

t,P

rim

eSp

ort

sSouth

wes

t,H

om

eSp

ort

sE

nte

rtain

men

t1.1

94

0.7

98

1.4

96

Fox

Sp

ort

sW

est,

Pri

me

Tic

ket

Old

,P

rim

eSp

ort

sW

est

2.0

04

0.5

78

3.4

67

MSG

Plu

s,F

oxSp

ort

sN

etN

ewY

ork

,Sp

ort

sChannel

New

York

1.2

18

0.5

98

2.0

37

Madis

on

Square

Gard

enN

etw

ork

MSG

2.2

72

0.5

95

3.8

16

Mid

-Atl

anti

cSp

ort

sN

etw

ork

,M

ASN

1.9

45

0.7

07

2.7

52

New

Engla

nd

Sp

ort

sN

etw

ork

5.6

78

0.8

00

7.0

95

Pri

me

Tic

ket

New

,F

oxSp

ort

sN

etW

est

21.2

95

0.6

06

2.1

36

Root

Sp

ort

sN

ort

hw

est,

Fox

Sp

ort

sN

ort

hw

est,

Nort

hw

est

Cable

Sp

ort

s,P

rim

eSp

ort

sN

ort

hw

est

1.6

57

0.6

05

2.7

39

Root

Sp

ort

sP

itts

burg

h,

Fox

Sp

ort

sP

itts

burg

h,

KB

L1.0

90

0.5

82

1.8

75

Root

Sp

ort

sR

ock

yM

ounta

in,

Fox

Sp

ort

sR

ock

yM

ounta

in,

Pri

me

Sp

ort

sR

ock

yM

ounta

in0.4

18

0.5

89

0.7

10

Sp

ort

sNet

New

York

1.2

10

0.6

06

1.9

96

Sun

Sp

ort

s,Sunsh

ine

Net

work

1.8

93

0.5

97

3.1

71

Yankee

sE

nte

rtain

men

tSp

ort

sY

ES

3.5

87

0.6

09

5.8

86

Notes:

This

table

pre

sente

des

tim

ate

dm

ean

month

lyw

illingnes

s-to

-pay

indollars

for

the

RSN

’s.

The

firs

tco

lum

nis

the

unco

ndit

ional

mea

n.

The

seco

nd

colu

mn

isth

efr

act

ion

of

consu

mer

sw

ith

posi

tive

valu

ati

ons.

The

thir

dco

lum

nis

the

mea

nam

ongst

those

wit

hp

osi

tive

valu

ati

ons.

46

The Welfare E ects of Vertical Integration in Multichannel ... · The Welfare E ects of Vertical...

Documents

Transcript of The Welfare E ects of Vertical Integration in Multichannel ... · The Welfare E ects of Vertical...